SlideShare a Scribd company logo

Broken Linux Performance Tools 2016

Download as PPTX, PDF
Brendan Gregg
Brendan Gregg

Brendan Gregg discusses the limitations and common misunderstandings of Linux performance tools in a talk at Scale14x. He emphasizes the need to verify metrics, understand their meanings, and avoid pitfalls in benchmarking and monitoring. The presentation covers various tools and metrics such as load averages, CPU usage, memory statistics, and disk I/O, highlighting discrepancies and providing recommendations for accurate performance assessment.

Technology
1 of 95
Downloaded 1,214 times
1
2
3
4
5
6
7
8
9
10
Most read
11
12
Most read
13
Most read
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
Broken Linux Performance Tools Brendan Gregg Senior Performance Architect, Netflix Jan 2016
Previously (SCaLE11x) Working Linux performance tools:
This Talk (SCaLE14x) Broken Linux performance tools: Objectives: – Bust assumptions about tools and metrics – Learn how to verify and find missing metrics – Avoid the common mistakes when benchmarking Note: Current software is discussed, which could be fixed in the future (by you!) BenchmarkingObservability
Broken Linux Performance Tools 2016
OBSERVABILITY Load Averages top %CPU iowait vmstat Overhead strace Java Profilers Monitoring
LOAD AVERAGES
Load Averages (1, 5, 15 min) • "load" – Usually CPU demand (run queue length/latency) – On Linux: CPU + uninterruptible I/O (e.g., disk) • "average" – Exponentially damped moving sum • "1, 5, and 15 minutes" – Constants used in the equation • Don't study these for longer than 10 seconds $ uptime 22:08:07 up 9:05, 1 user, load average: 11.42, 11.87, 12.12
t=0 Load begins (1 thread) 1 5 15 @ 1 min: 1 min avg =~ 0.62
TOP %CPU
top %CPU • Who is consuming CPU? • And by how much? $ top - 20:15:55 up 19:12, 1 user, load average: 7.96, 8.59, 7.05 Tasks: 470 total, 1 running, 468 sleeping, 0 stopped, 1 zombie %Cpu(s): 28.1 us, 0.4 sy, 0.0 ni, 71.2 id, 0.0 wa, 0.0 hi, 0.1 si, 0.1 st KiB Mem: 61663100 total, 61342588 used, 320512 free, 9544 buffers KiB Swap: 0 total, 0 used, 0 free. 3324696 cached Mem PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND 11959 apiprod 20 0 81.731g 0.053t 14476 S 935.8 92.1 13568:22 java 12595 snmp 20 0 21240 3256 1392 S 3.6 0.0 2:37.23 snmp-pass 10447 snmp 20 0 51512 6028 1432 S 2.0 0.0 2:12.12 snmpd 18463 apiprod 20 0 23932 1972 1176 R 0.7 0.0 0:00.07 top […]
top: Missing %CPU • Short-lived processes can be missing entirely – Process creates and exits in-between sampling /proc. e.g., software builds. – Try atop(1), or sampling using perf(1) • Short-lived processes may vanish on screen updates – I often use pidstat(1) on Linux instead, for concise scroll back
top: Misinterpreting %CPU • Different top(1)s use different calculations - On different OSes, check the man page, and run a test! • %CPU can mean: – A) Sum of per-CPU percents (0-Ncpu x 100%) consumed during the last interval – B) Percentage of total CPU capacity (0-100%) consumed during the last interval – C) (A) but historically damped (like load averages) – D) (B) " " "
top: %Cpu vs %CPU • This 4 CPU system is consuming: – 130% total CPU, via %Cpu(s) – 190% total CPU, via %CPU • Which one is right? Is either? – "A man with one watch knows the time; with two he's never sure" $ top - 15:52:58 up 10 days, 21:58, 2 users, load average: 0.27, 0.53, 0.41 Tasks: 180 total, 1 running, 179 sleeping, 0 stopped, 0 zombie %Cpu(s): 1.2 us, 24.5 sy, 0.0 ni, 67.2 id, 0.2 wa, 0.0 hi, 6.6 si, 0.4 st KiB Mem: 2872448 total, 2778160 used, 94288 free, 31424 buffers KiB Swap: 4151292 total, 76 used, 4151216 free. 2411728 cached Mem PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND 12678 root 20 0 96812 1100 912 S 100.4 0.0 0:23.52 iperf 12675 root 20 0 170544 1096 904 S 88.8 0.0 0:20.83 iperf 215 root 20 0 0 0 0 S 0.3 0.0 0:27.73 jbd2/sda1-8 […]
CPU Summary Statistics • %Cpu row is from /proc/stat • linux/Documentation/cpu-load.txt: • /proc/stat is used by everything for CPU stats In most cases the `/proc/stat' information reflects the reality quite closely, however due to the nature of how/when the kernel collects this data sometimes it can not be trusted at all.
%CPU
What is %CPU anyway? • "Good" %CPU: – Retiring instructions (provided they aren't a spin loop) – High IPC (Instructions-Per-Cycle) • "Bad" %CPU: – Stall cycles waiting on resources, usually memory I/O – Low IPC – Buying faster processors may make little difference • %CPU alone is ambiguous – Would love top(1) to split %CPU into cycles retiring vs stalled – Although, it gets worse…
CPU Speed Variation • Clock speed can vary thanks to: – Intel Turbo Boost: by hardware, based on power, temp, etc – Intel Speed Step: by software, controlled by the kernel • %CPU is still ambiguous, given IPC • Need to know the clock speed as well – 80% CPU (@3000MHz) != 4 x 20% CPU (@1600MHz) • CPU counters nowadays have "reference cycles" 80% CPU (1.6 IPC) 4 x 20% CPU (1.6 IPC) may not ==
Out-of-order Execution • CPUs execute uops out-of- order and in parallel across multiple functional units • %CPU doesn't account for how many units are active • Accounting each cycles as "stalled" or “retiring" is a simplification https://upload.wikimedia.org/wikipedia/commons/6/64/Intel_Nehalem_arch.svg
I/O WAIT
I/O Wait • Suggests system is disk I/O bound, but often misleading • Comparing I/O wait between system A and B: - higher might be bad: slower disks, more blocking - lower might be bad: slower processor and architecture consumes more CPU, obscuring I/O wait • Can be very useful when understood: another idle state $ mpstat -P ALL 1 08:06:43 PM CPU %usr %nice %sys %iowait %irq %soft %steal %guest %idle 08:06:44 PM all 53.45 0.00 3.77 0.00 0.00 0.39 0.13 0.00 42.26 […]
I/O Wait Venn Diagram "CPU" "I/O Wait""CPU" "Idle" CPU Waiting for disk I/O Per CPU:
FREE MEMORY
Free Memory • "free" is near-zero: I'm running out of memory! - No, it's in the file system cache, and is still free for apps to use • Linux free(1) explains it, but other tools, e.g. vmstat(1), don't • Some file systems (e.g., ZFS) may not be shown in the system's cached metrics at all www.linuxatemyram.com $ free -m total used free shared buffers cached Mem: 3750 1111 2639 0 147 527 -/+ buffers/cache: 436 3313 Swap: 0 0 0
VMSTAT
vmstat(1) • Linux: first line has some summary since boot values — confusing! • This system-wide summary is missing networking $ vmstat –Sm 1 procs -----------memory---------- ---swap-- -----io---- -system-- ----cpu---- r b swpd free buff cache si so bi bo in cs us sy id wa 8 0 0 1620 149 552 0 0 1 179 77 12 25 34 0 0 7 0 0 1598 149 552 0 0 0 0 205 186 46 13 0 0 8 0 0 1617 149 552 0 0 0 8 210 435 39 21 0 0 8 0 0 1589 149 552 0 0 0 0 218 219 42 17 0 0 […]
NETSTAT -S
netstat -s $ netstat -s Ip: 7962754 total packets received 8 with invalid addresses 0 forwarded 0 incoming packets discarded 7962746 incoming packets delivered 8019427 requests sent out Icmp: 382 ICMP messages received 0 input ICMP message failed. ICMP input histogram: destination unreachable: 125 timeout in transit: 257 3410 ICMP messages sent 0 ICMP messages failed ICMP output histogram: destination unreachable: 3410 IcmpMsg: InType3: 125 InType11: 257 OutType3: 3410 Tcp: 17337 active connections openings 395515 passive connection openings 8953 failed connection attempts 240214 connection resets received 3 connections established 7198375 segments received 7504939 segments send out 62696 segments retransmited 10 bad segments received. 1072 resets sent InCsumErrors: 5 Udp: 759925 packets received 3412 packets to unknown port received. 0 packet receive errors 784370 packets sent UdpLite: TcpExt: 858 invalid SYN cookies received 8951 resets received for embryonic SYN_RECV sockets 14 packets pruned from receive queue because of socket buffer overrun 6177 TCP sockets finished time wait in fast timer 293 packets rejects in established connections because of timestamp 733028 delayed acks sent 89 delayed acks further delayed because of locked socket Quick ack mode was activated 13214 times 336520 packets directly queued to recvmsg prequeue. 43964 packets directly received from backlog 11406012 packets directly received from prequeue 1039165 packets header predicted 7066 packets header predicted and directly queued to user 1428960 acknowledgments not containing data received 1004791 predicted acknowledgments 1 times recovered from packet loss due to fast retransmit 5044 times recovered from packet loss due to SACK data 2 bad SACKs received Detected reordering 4 times using SACK Detected reordering 11 times using time stamp 13 congestion windows fully recovered 11 congestion windows partially recovered using Hoe heuristic TCPDSACKUndo: 39 2384 congestion windows recovered after partial ack 228 timeouts after SACK recovery 100 timeouts in loss state 5018 fast retransmits 39 forward retransmits 783 retransmits in slow start 32455 other TCP timeouts TCPLossProbes: 30233 TCPLossProbeRecovery: 19070 992 sack retransmits failed 18 times receiver scheduled too late for direct processing 705 packets collapsed in receive queue due to low socket buffer 13658 DSACKs sent for old packets 8 DSACKs sent for out of order packets 13595 DSACKs received 33 DSACKs for out of order packets received 32 connections reset due to unexpected data 108 connections reset due to early user close 1608 connections aborted due to timeout TCPSACKDiscard: 4 TCPDSACKIgnoredOld: 1 TCPDSACKIgnoredNoUndo: 8649 TCPSpuriousRTOs: 445 TCPSackShiftFallback: 8588 TCPRcvCoalesce: 95854 TCPOFOQueue: 24741 TCPOFOMerge: 8 TCPChallengeACK: 1441 TCPSYNChallenge: 5 TCPSpuriousRtxHostQueues: 1 TCPAutoCorking: 4823 IpExt: InOctets: 1561561375 OutOctets: 1509416943 InNoECTPkts: 8201572 InECT1Pkts: 2 InECT0Pkts: 3844 InCEPkts: 306
netstat -s • Many metrics on Linux (can be over 200) • Still doesn't include everything: getting better, but don't assume everything is there • Includes typos & inconsistencies • Might be more readable to: cat /proc/net/snmp /proc/net/netstat • Totals since boot can be misleading • On Linux, -s needs -c support • Often no documentation outside kernel source code • Requires expertise to comprehend
DISK METRICS
Disk Metrics • All disk metrics are misleading • Disk %utilization / %busy – Logical devices (volume managers) and individual disks can process I/O in parallel, and may accept more I/O at 100% • Disk IOPS – High IOPS is "bad"? That depends… • Disk latency – Does it matter? File systems and volume managers try hard to hide latency and make it asynchronous – Better measuring latency via application->FS calls
FS CACHE METRICS
FS Cache Metrics • Size metrics exist: free -m • Activity metrics are missing: e.g., hit/miss ratio • Hacking stats using ftrace (/eBPF): # ./cachestat 1 Counting cache functions... Output every 1 seconds. HITS MISSES DIRTIES RATIO BUFFERS_MB CACHE_MB 210 869 0 19.5% 2 209 444 1413 0 23.9% 8 210 471 1399 0 25.2% 12 211 403 1507 3 21.1% 18 211 967 1853 3 34.3% 24 212 [...]
What You Can Do • Verify and understand existing metrics – Even %CPU can be misleading – Cross check with another tool & backend – Test with known workloads – Read the source, including comments – Use "known to be good" metrics to sanity test others • Find missing metrics – Follow the USE Method, and other methodologies – Draw a functional diagram • Burn it all down and start again from scratch?
PROFILERS
Linux perf • Can sample stack traces and summarize output: # perf report -n -stdio […] # Overhead Samples Command Shared Object Symbol # ........ ............ ....... ................. ............................. # 20.42% 605 bash [kernel.kallsyms] [k] xen_hypercall_xen_version | --- xen_hypercall_xen_version check_events | |--44.13%-- syscall_trace_enter | tracesys | | | |--35.58%-- __GI___libc_fcntl | | | | | |--65.26%-- do_redirection_internal | | | do_redirections | | | execute_builtin_or_function | | | execute_simple_command [… ~13,000 lines truncated …]
Too Much Output
… as a Flame Graph
PROFILER VISIBILITY
Java Profilers Java (+object stats) GC Kernel, libraries, JVM CPU Flame Graph
Java Profilers • Typical problems: – Sampling at safepoints (skew) – Method tracing observer effect – RUNNING != on-CPU (e.g., epoll) – Missing GC or JVM CPU time • Inaccurate (skewed) and incomplete profiles • Let's try a system profiler?
System Profilers with Java (x86) Java (missing stacks & symbols) Kernel TCP/IP GC Idle thread Time Locks epoll JVM compiler optimization #fail
COMPILER OPTIMIZATIONS
Broken System Stack Traces • Broken stacks (1 or 2 levels deep, junk values): • On x86 (x86_64), hotspot reuses the frame pointer register (RBP) as general purpose (a "compiler optimization"), which once upon a time made sense • gcc has -fno-omit-frame-pointer to avoid this – JDK8u60+ now has this as -XX:+PreserveFramePoiner # perf record –F 99 –a –g – sleep 30; perf script […] java 4579 cpu-clock: ffffffff8172adff tracesys ([kernel.kallsyms]) 7f4183bad7ce pthread_cond_timedwait@@GLIBC_2… java 4579 cpu-clock: 7f417908c10b [unknown] (/tmp/perf-4458.map) java 4579 cpu-clock: 7f4179101c97 [unknown] (/tmp/perf-4458.map)
• Missing symbols may show up as hex; e.g., Linux perf: • For applications, install debug symbol package • For JIT'd code, Linux perf already looks for an externally provided symbol file: /tmp/perf-PID.map – Find a way to do this for your runtime Missing Symbols # perf script Failed to open /tmp/perf-8131.map, continuing without symbols […] java 8131 cpu-clock: 7fff76f2dce1 [unknown] ([vdso]) 7fd3173f7a93 os::javaTimeMillis() (/usr/lib/jvm… 7fd301861e46 [unknown] (/tmp/perf-8131.map) […]
INSTRUCTION PROFILING
Instruction Profiling # perf annotate -i perf.data.noplooper --stdio Percent | Source code & Disassembly of noplooper ---------------------------------------------------- : Disassembly of section .text: : : 00000000004004ed <main>: 0.00 : 4004ed: push %rbp 0.00 : 4004ee: mov %rsp,%rbp 20.86 : 4004f1: nop 0.00 : 4004f2: nop 0.00 : 4004f3: nop 0.00 : 4004f4: nop 19.84 : 4004f5: nop 0.00 : 4004f6: nop 0.00 : 4004f7: nop 0.00 : 4004f8: nop 18.73 : 4004f9: nop 0.00 : 4004fa: nop 0.00 : 4004fb: nop 0.00 : 4004fc: nop 19.08 : 4004fd: nop 0.00 : 4004fe: nop 0.00 : 4004ff: nop 0.00 : 400500: nop 21.49 : 400501: jmp 4004f1 <main+0x4> • Often broken nowadays due to skid, out-of-order execution, and sampling the resumption instruction • Better with PEBS support
What You Can Do • Do stack trace profiling – Get stack traces to work – Get symbols to work – This all may be a lot of work. It's worth it! • Make CPU flame graphs!
OVERHEAD
tcpdump • Packet tracing doesn't scale. Overheads: – CPU cost of per-packet tracing (improved by [e]BPF) • Consider CPU budget per-packet at 10/40/100 GbE – Transfer to user-level (improved by ring buffers) – File system storage (more CPU, and disk I/O) – Possible additional network transfer • Can also drop packets when overloaded • You should only trace send/receive as a last resort – I solve problems by tracing lower frequency TCP events $ tcpdump -i eth0 -w /tmp/out.tcpdump tcpdump: listening on eth0, link-type EN10MB (Ethernet), capture size 65535 bytes ^C7985 packets captured 8996 packets received by filter 1010 packets dropped by kernel
STRACE
strace • Before: • After: • 442x slower. This is worst case. • strace(1) pauses the process twice for each syscall. This is like putting metering lights on your app. – "BUGS: A traced process runs slowly." – strace(1) man page $ dd if=/dev/zero of=/dev/null bs=1 count=500k […] 512000 bytes (512 kB) copied, 0.103851 s, 4.9 MB/s $ strace -eaccept dd if=/dev/zero of=/dev/null bs=1 count=500k […] 512000 bytes (512 kB) copied, 45.9599 s, 11.1 kB/s
PERF_EVENTS
perf_events • Buffered tracing helps, but you can still trace too much: • Overhead = event instrumentation cost X event frequency • Costs – Higher: event dumps (perf.data), stack traces, copyin/outs – Lower: counters, in-kernel aggregations (ftrace, eBPF) • Frequencies – Higher: instructions, scheduler, malloc/free, Java methods – Lower: process creation & destruction, disk I/O (usually) # perf record -e sched:sched_switch -a -g -- sleep 1 [ perf record: Woken up 3 times to write data ] [ perf record: Captured and wrote 100.212 MB perf.data (486550 samples) ]
VALGRIND
Valgrind • A suite of tools including an extensive leak detector • To its credit it does warn the end user "Your program will run much slower (eg. 20 to 30 times) than normal" – http://valgrind.org/docs/manual/quick-start.html
JAVA PROFILERS
Java Profilers • Some Java profilers have two modes: – Sampling stacks: eg, at 100 Hertz – Tracing methods: instrumenting and timing every method • Method timing has been described as "highly accurate", despite slowing the target by up to 1000x! • For more about Java profiler issues, see Nitsan Wakart's QCon2015 talk "Profilers are Lying Hobbitses"
What You Can Do • Understand how the profiler works – Measure overhead – Know the frequency of instrumented events • Use in-kernel summaries (ftrace, eBPF) – < 10,000 events/sec, probably ok – > 100,000 events/sec, overhead may start to be measurable
MONITORING
Monitoring • By now you should recognize these pathologies: – Let's just graph the system metrics! • That's not the problem that needs solving – Let's just trace everything and post process! • Now you have one million problems per second • Monitoring adds additional problems: – Let's have a cloud-wide dashboard update per-second! • From every instance? Packet overheads? – Now we have billions of metrics!
STATISTICS "Then there is the man who drowned crossing a stream with an average depth of six inches." – W.I.E. Gates
Statistics • Averages can be misleading – Hide latency outliers – Per-minute averages can hide multi-second issues • Percentiles can be misleading – Probability of hitting 99.9th latency may be more than 1/1000 after many dependency requests • Show the distribution: – Summarize: histogram, density plot, frequency trail – Over-time: scatter plot, heat map
Average Latency • When the index of central tendency isn't…
VISUALIZATIONS
Traffic Lights RED == bad, GREEN == good …misleading for subjective metrics Better suited for objective metrics
Tachometers …especially with arbitrary color highlighting
Pie Charts …for real-time metrics usr sys wait idle
What You Can Do • Monitoring: – Verify metrics, test overhead (same as tools) • Statistics: – Ask how is this calculated? – Study the full distribution • Visualizations: – Use histograms, heat maps, flame graphs
BENCHMARKING Benchmarks Common Mistakes Micro Macro Kitchen-Sink bonnie++ Apache Bench
BENCHMARKS
~100% of Benchmarks are Wrong • "Most popular benchmarks are flawed" – Traeger, A., E. Zadok, N. Joukov, and C. Wright. "A Nine Year Study of File System and Storage Benchmarking," ACM Transactions on Storage, 2008. • All alternates can also be flawed
COMMON MISTAKES
Common Mistakes 1. Testing the wrong target – eg, FS cache instead of disk; misconfiguration 2. Choosing the wrong target – eg, disk instead of FS cache … doesn’t resemble real world 3. Invalid results – benchmark software bugs 4. Ignoring errors – error path may be fast! 5. Ignoring variance or perturbations – real workload isn't steady/consistent, which matters 6. Misleading results – Casual benchmarking: you benchmark A, but actually measure B, and conclude you measured C
MICRO BENCHMARKS
Micro Benchmarks • Test a specific function in isolation. e.g.: – File system maximum cached read ops/sec – Network maximum throughput • Examples of bad microbenchmarks: – gitpid() in a tight loop – speed of /dev/zero and /dev/null • Common problems: – Testing a workload that is not very relevant – Missing other workloads that are relevant
MACRO BENCHMARKS
Macro Benchmarks • Simulate application user load. e.g.: – Simulated web client transaction • Common problems: – Misplaced trust: believed to be realistic, but misses variance, errors, perturbations, etc. – Complex to debug, verify, and root cause
KITCHEN SINK BENCHMARKS
Kitchen Sink Benchmarks • Run everything! – Mostly random benchmarks found on the Internet, where most are are broken or irrelevant – Developers focus on collecting more benchmarks than verifying or fixing the existing ones • Myth that more benchmarks == greater accuracy – No, use active benchmarking (analysis)
BONNIE++
bonnie++ • "simple tests of hard drive and file system performance" • First metric printed: per character sequential output • What I found it actually tested: – 1 byte writes to libc (via putc()) – 4 Kbyte writes from libc -> FS (depends on OS; see setbuffer()) – 128 Kbyte async writes to disk (depends on storage stack) – Any file system throttles that may be present (eg, ionice) – C++ code, to some extent (bonnie++ 10% slower than Bonnie) • Actual limiter: – Single threaded write_block_putc() and putc() calls • Now thankfully fixed
APACHE BENCH
Apache Bench • HTTP web server benchmark • Single thread limited (use wrk for multi-threaded) • Keep-alive option (-k): – without: Can become an unrealistic TCP session benchmark – with: Can become an unrealistic server throughput test • Performance issues of ab's own code
UNIXBENCH
UnixBench • The original kitchen-sink micro benchmark from 1984, published in BYTE magazine • Results summarized as "The BYTE Index". Including: • Many problems, starting with… system: dhry2reg Dhrystone 2 using register variables whetstone-double Double-Precision Whetstone syscall System Call Overhead pipe Pipe Throughput context1 Pipe-based Context Switching spawn Process Creation execl Execl Throughput fstime-w File Write 1024 bufsize 2000 maxblocks fstime-r File Read 1024 bufsize 2000 maxblocks fstime File Copy 1024 bufsize 2000 maxblocks fsbuffer-w File Write 256 bufsize 500 maxblocks fsbuffer-r File Read 256 bufsize 500 maxblocks fsbuffer File Copy 256 bufsize 500 maxblocks fsdisk-w File Write 4096 bufsize 8000 maxblocks […]
UnixBench Makefile • Default (by ./Run) for Linux. Would you edit it? Then what? • I "fixed" it and "improved" Dhrystone 2 performance by 64% ## Very generic #OPTON = -O ## For Linux 486/Pentium, GCC 2.7.x and 2.8.x #OPTON = -O2 -fomit-frame-pointer -fforce-addr -fforce-mem -ffast-math # -m486 -malign-loops=2 -malign-jumps=2 -malign-functions=2 ## For Linux, GCC previous to 2.7.0 #OPTON = -O2 -fomit-frame-pointer -fforce-addr -fforce-mem -ffast-math -m486 #OPTON = -O2 -fomit-frame-pointer -fforce-addr -fforce-mem -ffast-math # -m386 -malign-loops=1 -malign-jumps=1 -malign-functions=1 ## For Solaris 2, or general-purpose GCC 2.7.x OPTON = -O2 -fomit-frame-pointer -fforce-addr -ffast-math -Wall ## For Digital Unix v4.x, with DEC cc v5.x #OPTON = -O4 #CFLAGS = -DTIME -std1 -verbose -w0
UnixBench Documentation "The results will depend not only on your hardware, but on your operating system, libraries, and even compiler." "So you may want to make sure that all your test systems are running the same version of the OS; or at least publish the OS and compuiler versions with your results." … UnixBench was innovative & useful, but it's time has passed
What You Can Do • Match the benchmark to your workload • Active Benchmarking 1. Configure the benchmark to run in steady state, 24x7 2. Do root-cause analysis of benchmark performance 3. Answer: why X and not 10X? Limiting factor? It can take 1-2 weeks to debug a single benchmark
Summary
Observe Everything • Trust nothing. Verify. Write small tests. • Pose Q's first then find the metrics. e.g., functional diagrams: Reference: http://www.brendangregg.com/linuxperf.html
• e.g., Java Mixed-Mode Flame Graphs: Profile Everything Java JVM Kernel GC Reference: http://www.brendangregg.com/linuxperf.html
Visualize Everything • Full distributions of latency. e.g., heat maps: Reference: http://queue.acm.org/detail.cfm?id=1809426
Benchmark Nothing! (if you must, use Active Benchmarking)
Links & References • Things that aren't broken: – http://www.brendangregg.com/linuxperf.html • References: – https://upload.wikimedia.org/wikipedia/commons/6/64/Intel_Nehalem_arch.svg – http://www.linuxatemyram.com/ – Traeger, A., E. Zadok, N. Joukov, and C. Wright. “A Nine Year Study of File System and Storage Benchmarking,” ACM Trans- actions on Storage, 2008. – http://www.brendangregg.com/blog/2014-06-09/java-cpu-sampling-using-hprof.html – http://www.brendangregg.com/activebenchmarking.html – https://blogs.oracle.com/roch/entry/decoding_bonnie – http://www.brendangregg.com/blog/2014-05-02/compilers-love-messing-with- benchmarks.html – https://code.google.com/p/byte-unixbench/ – https://qconsf.com/sf2015/presentation/how-not-measure-latency – https://qconsf.com/system/files/presentation-slides/profilers_are_lying_hobbitses.pdf – Caution signs drawn be me, inspired by real-world signs
Thanks • Questions? • http://techblog.netflix.com • http://slideshare.net/brendangregg • http://www.brendangregg.com • bgregg@netflix.com • @brendangregg Jan 2016

More Related Content

What's hot (20)

PDF
Java Performance Analysis on Linux with Flame Graphs
Brendan Gregg
 
PDF
Linux Performance Analysis and Tools
Brendan Gregg
 
PDF
USENIX ATC 2017: Visualizing Performance with Flame Graphs
Brendan Gregg
 
PDF
New Ways to Find Latency in Linux Using Tracing
ScyllaDB
 
PPTX
Linux Memory Management
Ni Zo-Ma
 
PDF
BPF - in-kernel virtual machine
Alexei Starovoitov
 
PDF
Linux Systems Performance 2016
Brendan Gregg
 
PDF
Kafka’s New Control Plane: The Quorum Controller | Colin McCabe, Confluent
HostedbyConfluent
 
PDF
Linux kernel tracing
Viller Hsiao
 
ODP
Continguous Memory Allocator in the Linux Kernel
Kernel TLV
 
PDF
BPF: Tracing and more
Brendan Gregg
 
PDF
Blazing Performance with Flame Graphs
Brendan Gregg
 
PDF
Velocity 2017 Performance analysis superpowers with Linux eBPF
Brendan Gregg
 
PPTX
Linux kernel debugging
Hao-Ran Liu
 
PDF
re:Invent 2019 BPF Performance Analysis at Netflix
Brendan Gregg
 
PDF
Linux 4.x Tracing Tools: Using BPF Superpowers
Brendan Gregg
 
PPTX
Staring into the eBPF Abyss
Sasha Goldshtein
 
PDF
Meet cute-between-ebpf-and-tracing
Viller Hsiao
 
PDF
Memory Mapping Implementation (mmap) in Linux Kernel
Adrian Huang
 
PPTX
Slab Allocator in Linux Kernel
Adrian Huang
 
Java Performance Analysis on Linux with Flame Graphs
Brendan Gregg
 
Linux Performance Analysis and Tools
Brendan Gregg
 
USENIX ATC 2017: Visualizing Performance with Flame Graphs
Brendan Gregg
 
New Ways to Find Latency in Linux Using Tracing
ScyllaDB
 
Linux Memory Management
Ni Zo-Ma
 
BPF - in-kernel virtual machine
Alexei Starovoitov
 
Linux Systems Performance 2016
Brendan Gregg
 
Kafka’s New Control Plane: The Quorum Controller | Colin McCabe, Confluent
HostedbyConfluent
 
Linux kernel tracing
Viller Hsiao
 
Continguous Memory Allocator in the Linux Kernel
Kernel TLV
 
BPF: Tracing and more
Brendan Gregg
 
Blazing Performance with Flame Graphs
Brendan Gregg
 
Velocity 2017 Performance analysis superpowers with Linux eBPF
Brendan Gregg
 
Linux kernel debugging
Hao-Ran Liu
 
re:Invent 2019 BPF Performance Analysis at Netflix
Brendan Gregg
 
Linux 4.x Tracing Tools: Using BPF Superpowers
Brendan Gregg
 
Staring into the eBPF Abyss
Sasha Goldshtein
 
Meet cute-between-ebpf-and-tracing
Viller Hsiao
 
Memory Mapping Implementation (mmap) in Linux Kernel
Adrian Huang
 
Slab Allocator in Linux Kernel
Adrian Huang
 

Viewers also liked (19)

POTX
Performance Tuning EC2 Instances
Brendan Gregg
 
PDF
Systems Performance: Enterprise and the Cloud
Brendan Gregg
 
PDF
LISA17 Container Performance Analysis
Brendan Gregg
 
PDF
Monitorama 2015 Netflix Instance Analysis
Brendan Gregg
 
PDF
Linux network stack
Takuya ASADA
 
PDF
Kernel Recipes 2017 - Using Linux perf at Netflix - Brendan Gregg
Anne Nicolas
 
PPT
Mian Nawaz Sharif PMLN
Huawei Technologies
 
PPTX
Unique ID generation in distributed systems
Dave Gardner
 
PPTX
No data loss pipeline with apache kafka
Jiangjie Qin
 
PDF
ACM DEBS 2015: Realtime Streaming Analytics Patterns
Srinath Perera
 
PDF
Docker в Badoo: ПМЖ или временная регистрация
Badoo Development
 
PDF
Docker networking
Badoo Development
 
PDF
Архитектура поиска в Avito / Андрей Смирнов (Avito)
Ontico
 
PDF
Docker penetration
Badoo Development
 
PDF
Мониторь, автоматизируй Docker
Badoo Development
 
PDF
Database as a Service on the Oracle Database Appliance Platform
Maris Elsins
 
PDF
Oracle Databases on AWS - Getting the Best Out of RDS and EC2
Maris Elsins
 
PDF
Oracle RAC Internals - The Cache Fusion Edition
Markus Michalewicz
 
PDF
Cloud Native Camel Design Patterns
Bilgin Ibryam
 
Performance Tuning EC2 Instances
Brendan Gregg
 
Systems Performance: Enterprise and the Cloud
Brendan Gregg
 
LISA17 Container Performance Analysis
Brendan Gregg
 
Monitorama 2015 Netflix Instance Analysis
Brendan Gregg
 
Linux network stack
Takuya ASADA
 
Kernel Recipes 2017 - Using Linux perf at Netflix - Brendan Gregg
Anne Nicolas
 
Mian Nawaz Sharif PMLN
Huawei Technologies
 
Unique ID generation in distributed systems
Dave Gardner
 
No data loss pipeline with apache kafka
Jiangjie Qin
 
ACM DEBS 2015: Realtime Streaming Analytics Patterns
Srinath Perera
 
Docker в Badoo: ПМЖ или временная регистрация
Badoo Development
 
Docker networking
Badoo Development
 
Архитектура поиска в Avito / Андрей Смирнов (Avito)
Ontico
 
Docker penetration
Badoo Development
 
Мониторь, автоматизируй Docker
Badoo Development
 
Database as a Service on the Oracle Database Appliance Platform
Maris Elsins
 
Oracle Databases on AWS - Getting the Best Out of RDS and EC2
Maris Elsins
 
Oracle RAC Internals - The Cache Fusion Edition
Markus Michalewicz
 
Cloud Native Camel Design Patterns
Bilgin Ibryam
 
Ad

Similar to Broken Linux Performance Tools 2016 (20)

PPTX
QCon 2015 Broken Performance Tools
Brendan Gregg
 
PDF
Broken Performance Tools
C4Media
 
PDF
YOW2020 Linux Systems Performance
Brendan Gregg
 
PDF
OSDC 2015: Georg Schönberger | Linux Performance Profiling and Monitoring
NETWAYS
 
PDF
Linux Performance Profiling and Monitoring
Georg Schönberger
 
PDF
Linux Performance Tools
Brendan Gregg
 
PDF
Linux Performance Tools 2014
Brendan Gregg
 
PDF
OSMC 2015: Linux Performance Profiling and Monitoring by Werner Fischer
NETWAYS
 
PDF
OSMC 2015 | Linux Performance Profiling and Monitoring by Werner Fischer
NETWAYS
 
PDF
OSDC 2017 - Werner Fischer - Linux performance profiling and monitoring
NETWAYS
 
PDF
MeetBSD2014 Performance Analysis
Brendan Gregg
 
PDF
test
WentingLiu4
 
PPT
Servers and Processes: Behavior and Analysis
dreamwidth
 
PDF
Analyzing OS X Systems Performance with the USE Method
Brendan Gregg
 
PDF
linux monitoring and performance tunning
iman darabi
 
PDF
200.1,2-Capacity Planning
behrad eslamifar
 
PDF
Kernel Recipes 2019 - Metrics are money
Anne Nicolas
 
PPT
Linux monitoring and Troubleshooting for DBA's
Mydbops
 
PDF
Essential Linux Commands for DBAs
Gokhan Atil
 
PPTX
Linux optimization strategy plan by shiv
thedatabasearchitect
 
QCon 2015 Broken Performance Tools
Brendan Gregg
 
Broken Performance Tools
C4Media
 
YOW2020 Linux Systems Performance
Brendan Gregg
 
OSDC 2015: Georg Schönberger | Linux Performance Profiling and Monitoring
NETWAYS
 
Linux Performance Profiling and Monitoring
Georg Schönberger
 
Linux Performance Tools
Brendan Gregg
 
Linux Performance Tools 2014
Brendan Gregg
 
OSMC 2015: Linux Performance Profiling and Monitoring by Werner Fischer
NETWAYS
 
OSMC 2015 | Linux Performance Profiling and Monitoring by Werner Fischer
NETWAYS
 
OSDC 2017 - Werner Fischer - Linux performance profiling and monitoring
NETWAYS
 
MeetBSD2014 Performance Analysis
Brendan Gregg
 
test
WentingLiu4
 
Servers and Processes: Behavior and Analysis
dreamwidth
 
Analyzing OS X Systems Performance with the USE Method
Brendan Gregg
 
linux monitoring and performance tunning
iman darabi
 
200.1,2-Capacity Planning
behrad eslamifar
 
Kernel Recipes 2019 - Metrics are money
Anne Nicolas
 
Linux monitoring and Troubleshooting for DBA's
Mydbops
 
Essential Linux Commands for DBAs
Gokhan Atil
 
Linux optimization strategy plan by shiv
thedatabasearchitect
 
Ad

More from Brendan Gregg (20)

PDF
YOW2021 Computing Performance
Brendan Gregg
 
PDF
IntelON 2021 Processor Benchmarking
Brendan Gregg
 
PDF
Performance Wins with eBPF: Getting Started (2021)
Brendan Gregg
 
PDF
Systems@Scale 2021 BPF Performance Getting Started
Brendan Gregg
 
PDF
Computing Performance: On the Horizon (2021)
Brendan Gregg
 
PDF
BPF Internals (eBPF)
Brendan Gregg
 
PDF
Performance Wins with BPF: Getting Started
Brendan Gregg
 
PDF
UM2019 Extended BPF: A New Type of Software
Brendan Gregg
 
PDF
LPC2019 BPF Tracing Tools
Brendan Gregg
 
PDF
LSFMM 2019 BPF Observability
Brendan Gregg
 
PDF
YOW2018 CTO Summit: Working at netflix
Brendan Gregg
 
PDF
eBPF Perf Tools 2019
Brendan Gregg
 
PDF
YOW2018 Cloud Performance Root Cause Analysis at Netflix
Brendan Gregg
 
PDF
BPF Tools 2017
Brendan Gregg
 
PDF
NetConf 2018 BPF Observability
Brendan Gregg
 
PDF
FlameScope 2018
Brendan Gregg
 
PDF
ATO Linux Performance 2018
Brendan Gregg
 
PDF
Linux Performance 2018 (PerconaLive keynote)
Brendan Gregg
 
PDF
How Netflix Tunes EC2 Instances for Performance
Brendan Gregg
 
PDF
Kernel Recipes 2017: Performance Analysis with BPF
Brendan Gregg
 
YOW2021 Computing Performance
Brendan Gregg
 
IntelON 2021 Processor Benchmarking
Brendan Gregg
 
Performance Wins with eBPF: Getting Started (2021)
Brendan Gregg
 
Systems@Scale 2021 BPF Performance Getting Started
Brendan Gregg
 
Computing Performance: On the Horizon (2021)
Brendan Gregg
 
BPF Internals (eBPF)
Brendan Gregg
 
Performance Wins with BPF: Getting Started
Brendan Gregg
 
UM2019 Extended BPF: A New Type of Software
Brendan Gregg
 
LPC2019 BPF Tracing Tools
Brendan Gregg
 
LSFMM 2019 BPF Observability
Brendan Gregg
 
YOW2018 CTO Summit: Working at netflix
Brendan Gregg
 
eBPF Perf Tools 2019
Brendan Gregg
 
YOW2018 Cloud Performance Root Cause Analysis at Netflix
Brendan Gregg
 
BPF Tools 2017
Brendan Gregg
 
NetConf 2018 BPF Observability
Brendan Gregg
 
FlameScope 2018
Brendan Gregg
 
ATO Linux Performance 2018
Brendan Gregg
 
Linux Performance 2018 (PerconaLive keynote)
Brendan Gregg
 
How Netflix Tunes EC2 Instances for Performance
Brendan Gregg
 
Kernel Recipes 2017: Performance Analysis with BPF
Brendan Gregg
 

Recently uploaded (20)

PPTX
AI Penetration Testing Essentials: A Cybersecurity Guide for 2025
defencerabbit
 
PDF
The 2025 InfraRed Report - Redpoint Ventures
Razin Mustafiz
 
PPT
Ericsson LTE presentation SEMINAR 2010.ppt
npat3
 
PPTX
Digital Circuits, important subject in CS
contactparinay1
 
DOCX
Python coding for beginners !! Start now!#
Rajni Bhardwaj Grover
 
PPTX
COMPARISON OF RASTER ANALYSIS TOOLS OF QGIS AND ARCGIS
Sharanya Sarkar
 
PDF
UiPath DevConnect 2025: Agentic Automation Community User Group Meeting
DianaGray10
 
PPTX
Seamless Tech Experiences Showcasing Cross-Platform App Design.pptx
presentifyai
 
PDF
SIZING YOUR AIR CONDITIONER---A PRACTICAL GUIDE.pdf
Muhammad Rizwan Akram
 
PDF
Kit-Works Team Study_20250627_한달만에만든사내서비스키링(양다윗).pdf
Wonjun Hwang
 
PDF
Staying Human in a Machine- Accelerated World
Catalin Jora
 
PDF
[Newgen] NewgenONE Marvin Brochure 1.pdf
darshakparmar
 
PPTX
Mastering ODC + Okta Configuration - Chennai OSUG
HathiMaryA
 
PDF
AI Agents in the Cloud: The Rise of Agentic Cloud Architecture
Lilly Gracia
 
PPTX
The Project Compass - GDG on Campus MSIT
dscmsitkol
 
PDF
Peak of Data & AI Encore AI-Enhanced Workflows for the Real World
Safe Software
 
PDF
Bitcoin for Millennials podcast with Bram, Power Laws of Bitcoin
Stephen Perrenod
 
PDF
Newgen Beyond Frankenstein_Build vs Buy_Digital_version.pdf
darshakparmar
 
PDF
What’s my job again? Slides from Mark Simos talk at 2025 Tampa BSides
Mark Simos
 
PDF
Newgen 2022-Forrester Newgen TEI_13 05 2022-The-Total-Economic-Impact-Newgen-...
darshakparmar
 
AI Penetration Testing Essentials: A Cybersecurity Guide for 2025
defencerabbit
 
The 2025 InfraRed Report - Redpoint Ventures
Razin Mustafiz
 
Ericsson LTE presentation SEMINAR 2010.ppt
npat3
 
Digital Circuits, important subject in CS
contactparinay1
 
Python coding for beginners !! Start now!#
Rajni Bhardwaj Grover
 
COMPARISON OF RASTER ANALYSIS TOOLS OF QGIS AND ARCGIS
Sharanya Sarkar
 
UiPath DevConnect 2025: Agentic Automation Community User Group Meeting
DianaGray10
 
Seamless Tech Experiences Showcasing Cross-Platform App Design.pptx
presentifyai
 
SIZING YOUR AIR CONDITIONER---A PRACTICAL GUIDE.pdf
Muhammad Rizwan Akram
 
Kit-Works Team Study_20250627_한달만에만든사내서비스키링(양다윗).pdf
Wonjun Hwang
 
Staying Human in a Machine- Accelerated World
Catalin Jora
 
[Newgen] NewgenONE Marvin Brochure 1.pdf
darshakparmar
 
Mastering ODC + Okta Configuration - Chennai OSUG
HathiMaryA
 
AI Agents in the Cloud: The Rise of Agentic Cloud Architecture
Lilly Gracia
 
The Project Compass - GDG on Campus MSIT
dscmsitkol
 
Peak of Data & AI Encore AI-Enhanced Workflows for the Real World
Safe Software
 
Bitcoin for Millennials podcast with Bram, Power Laws of Bitcoin
Stephen Perrenod
 
Newgen Beyond Frankenstein_Build vs Buy_Digital_version.pdf
darshakparmar
 
What’s my job again? Slides from Mark Simos talk at 2025 Tampa BSides
Mark Simos
 
Newgen 2022-Forrester Newgen TEI_13 05 2022-The-Total-Economic-Impact-Newgen-...
darshakparmar
 

Broken Linux Performance Tools 2016

  • 1. Broken Linux Performance Tools Brendan Gregg Senior Performance Architect, Netflix Jan 2016
  • 3. This Talk (SCaLE14x) Broken Linux performance tools: Objectives: – Bust assumptions about tools and metrics – Learn how to verify and find missing metrics – Avoid the common mistakes when benchmarking Note: Current software is discussed, which could be fixed in the future (by you!) BenchmarkingObservability
  • 5. OBSERVABILITY Load Averages top %CPU iowait vmstat Overhead strace Java Profilers Monitoring
  • 7. Load Averages (1, 5, 15 min) • "load" – Usually CPU demand (run queue length/latency) – On Linux: CPU + uninterruptible I/O (e.g., disk) • "average" – Exponentially damped moving sum • "1, 5, and 15 minutes" – Constants used in the equation • Don't study these for longer than 10 seconds $ uptime 22:08:07 up 9:05, 1 user, load average: 11.42, 11.87, 12.12
  • 8. t=0 Load begins (1 thread) 1 5 15 @ 1 min: 1 min avg =~ 0.62
  • 10. top %CPU • Who is consuming CPU? • And by how much? $ top - 20:15:55 up 19:12, 1 user, load average: 7.96, 8.59, 7.05 Tasks: 470 total, 1 running, 468 sleeping, 0 stopped, 1 zombie %Cpu(s): 28.1 us, 0.4 sy, 0.0 ni, 71.2 id, 0.0 wa, 0.0 hi, 0.1 si, 0.1 st KiB Mem: 61663100 total, 61342588 used, 320512 free, 9544 buffers KiB Swap: 0 total, 0 used, 0 free. 3324696 cached Mem PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND 11959 apiprod 20 0 81.731g 0.053t 14476 S 935.8 92.1 13568:22 java 12595 snmp 20 0 21240 3256 1392 S 3.6 0.0 2:37.23 snmp-pass 10447 snmp 20 0 51512 6028 1432 S 2.0 0.0 2:12.12 snmpd 18463 apiprod 20 0 23932 1972 1176 R 0.7 0.0 0:00.07 top […]
  • 11. top: Missing %CPU • Short-lived processes can be missing entirely – Process creates and exits in-between sampling /proc. e.g., software builds. – Try atop(1), or sampling using perf(1) • Short-lived processes may vanish on screen updates – I often use pidstat(1) on Linux instead, for concise scroll back
  • 12. top: Misinterpreting %CPU • Different top(1)s use different calculations - On different OSes, check the man page, and run a test! • %CPU can mean: – A) Sum of per-CPU percents (0-Ncpu x 100%) consumed during the last interval – B) Percentage of total CPU capacity (0-100%) consumed during the last interval – C) (A) but historically damped (like load averages) – D) (B) " " "
  • 13. top: %Cpu vs %CPU • This 4 CPU system is consuming: – 130% total CPU, via %Cpu(s) – 190% total CPU, via %CPU • Which one is right? Is either? – "A man with one watch knows the time; with two he's never sure" $ top - 15:52:58 up 10 days, 21:58, 2 users, load average: 0.27, 0.53, 0.41 Tasks: 180 total, 1 running, 179 sleeping, 0 stopped, 0 zombie %Cpu(s): 1.2 us, 24.5 sy, 0.0 ni, 67.2 id, 0.2 wa, 0.0 hi, 6.6 si, 0.4 st KiB Mem: 2872448 total, 2778160 used, 94288 free, 31424 buffers KiB Swap: 4151292 total, 76 used, 4151216 free. 2411728 cached Mem PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND 12678 root 20 0 96812 1100 912 S 100.4 0.0 0:23.52 iperf 12675 root 20 0 170544 1096 904 S 88.8 0.0 0:20.83 iperf 215 root 20 0 0 0 0 S 0.3 0.0 0:27.73 jbd2/sda1-8 […]
  • 14. CPU Summary Statistics • %Cpu row is from /proc/stat • linux/Documentation/cpu-load.txt: • /proc/stat is used by everything for CPU stats In most cases the `/proc/stat' information reflects the reality quite closely, however due to the nature of how/when the kernel collects this data sometimes it can not be trusted at all.
  • 15. %CPU
  • 16. What is %CPU anyway? • "Good" %CPU: – Retiring instructions (provided they aren't a spin loop) – High IPC (Instructions-Per-Cycle) • "Bad" %CPU: – Stall cycles waiting on resources, usually memory I/O – Low IPC – Buying faster processors may make little difference • %CPU alone is ambiguous – Would love top(1) to split %CPU into cycles retiring vs stalled – Although, it gets worse…
  • 17. CPU Speed Variation • Clock speed can vary thanks to: – Intel Turbo Boost: by hardware, based on power, temp, etc – Intel Speed Step: by software, controlled by the kernel • %CPU is still ambiguous, given IPC • Need to know the clock speed as well – 80% CPU (@3000MHz) != 4 x 20% CPU (@1600MHz) • CPU counters nowadays have "reference cycles" 80% CPU (1.6 IPC) 4 x 20% CPU (1.6 IPC) may not ==
  • 18. Out-of-order Execution • CPUs execute uops out-of- order and in parallel across multiple functional units • %CPU doesn't account for how many units are active • Accounting each cycles as "stalled" or “retiring" is a simplification https://upload.wikimedia.org/wikipedia/commons/6/64/Intel_Nehalem_arch.svg
  • 20. I/O Wait • Suggests system is disk I/O bound, but often misleading • Comparing I/O wait between system A and B: - higher might be bad: slower disks, more blocking - lower might be bad: slower processor and architecture consumes more CPU, obscuring I/O wait • Can be very useful when understood: another idle state $ mpstat -P ALL 1 08:06:43 PM CPU %usr %nice %sys %iowait %irq %soft %steal %guest %idle 08:06:44 PM all 53.45 0.00 3.77 0.00 0.00 0.39 0.13 0.00 42.26 […]
  • 21. I/O Wait Venn Diagram "CPU" "I/O Wait""CPU" "Idle" CPU Waiting for disk I/O Per CPU:
  • 23. Free Memory • "free" is near-zero: I'm running out of memory! - No, it's in the file system cache, and is still free for apps to use • Linux free(1) explains it, but other tools, e.g. vmstat(1), don't • Some file systems (e.g., ZFS) may not be shown in the system's cached metrics at all www.linuxatemyram.com $ free -m total used free shared buffers cached Mem: 3750 1111 2639 0 147 527 -/+ buffers/cache: 436 3313 Swap: 0 0 0
  • 25. vmstat(1) • Linux: first line has some summary since boot values — confusing! • This system-wide summary is missing networking $ vmstat –Sm 1 procs -----------memory---------- ---swap-- -----io---- -system-- ----cpu---- r b swpd free buff cache si so bi bo in cs us sy id wa 8 0 0 1620 149 552 0 0 1 179 77 12 25 34 0 0 7 0 0 1598 149 552 0 0 0 0 205 186 46 13 0 0 8 0 0 1617 149 552 0 0 0 8 210 435 39 21 0 0 8 0 0 1589 149 552 0 0 0 0 218 219 42 17 0 0 […]
  • 27. netstat -s $ netstat -s Ip: 7962754 total packets received 8 with invalid addresses 0 forwarded 0 incoming packets discarded 7962746 incoming packets delivered 8019427 requests sent out Icmp: 382 ICMP messages received 0 input ICMP message failed. ICMP input histogram: destination unreachable: 125 timeout in transit: 257 3410 ICMP messages sent 0 ICMP messages failed ICMP output histogram: destination unreachable: 3410 IcmpMsg: InType3: 125 InType11: 257 OutType3: 3410 Tcp: 17337 active connections openings 395515 passive connection openings 8953 failed connection attempts 240214 connection resets received 3 connections established 7198375 segments received 7504939 segments send out 62696 segments retransmited 10 bad segments received. 1072 resets sent InCsumErrors: 5 Udp: 759925 packets received 3412 packets to unknown port received. 0 packet receive errors 784370 packets sent UdpLite: TcpExt: 858 invalid SYN cookies received 8951 resets received for embryonic SYN_RECV sockets 14 packets pruned from receive queue because of socket buffer overrun 6177 TCP sockets finished time wait in fast timer 293 packets rejects in established connections because of timestamp 733028 delayed acks sent 89 delayed acks further delayed because of locked socket Quick ack mode was activated 13214 times 336520 packets directly queued to recvmsg prequeue. 43964 packets directly received from backlog 11406012 packets directly received from prequeue 1039165 packets header predicted 7066 packets header predicted and directly queued to user 1428960 acknowledgments not containing data received 1004791 predicted acknowledgments 1 times recovered from packet loss due to fast retransmit 5044 times recovered from packet loss due to SACK data 2 bad SACKs received Detected reordering 4 times using SACK Detected reordering 11 times using time stamp 13 congestion windows fully recovered 11 congestion windows partially recovered using Hoe heuristic TCPDSACKUndo: 39 2384 congestion windows recovered after partial ack 228 timeouts after SACK recovery 100 timeouts in loss state 5018 fast retransmits 39 forward retransmits 783 retransmits in slow start 32455 other TCP timeouts TCPLossProbes: 30233 TCPLossProbeRecovery: 19070 992 sack retransmits failed 18 times receiver scheduled too late for direct processing 705 packets collapsed in receive queue due to low socket buffer 13658 DSACKs sent for old packets 8 DSACKs sent for out of order packets 13595 DSACKs received 33 DSACKs for out of order packets received 32 connections reset due to unexpected data 108 connections reset due to early user close 1608 connections aborted due to timeout TCPSACKDiscard: 4 TCPDSACKIgnoredOld: 1 TCPDSACKIgnoredNoUndo: 8649 TCPSpuriousRTOs: 445 TCPSackShiftFallback: 8588 TCPRcvCoalesce: 95854 TCPOFOQueue: 24741 TCPOFOMerge: 8 TCPChallengeACK: 1441 TCPSYNChallenge: 5 TCPSpuriousRtxHostQueues: 1 TCPAutoCorking: 4823 IpExt: InOctets: 1561561375 OutOctets: 1509416943 InNoECTPkts: 8201572 InECT1Pkts: 2 InECT0Pkts: 3844 InCEPkts: 306
  • 28. netstat -s • Many metrics on Linux (can be over 200) • Still doesn't include everything: getting better, but don't assume everything is there • Includes typos & inconsistencies • Might be more readable to: cat /proc/net/snmp /proc/net/netstat • Totals since boot can be misleading • On Linux, -s needs -c support • Often no documentation outside kernel source code • Requires expertise to comprehend
  • 30. Disk Metrics • All disk metrics are misleading • Disk %utilization / %busy – Logical devices (volume managers) and individual disks can process I/O in parallel, and may accept more I/O at 100% • Disk IOPS – High IOPS is "bad"? That depends… • Disk latency – Does it matter? File systems and volume managers try hard to hide latency and make it asynchronous – Better measuring latency via application->FS calls
  • 32. FS Cache Metrics • Size metrics exist: free -m • Activity metrics are missing: e.g., hit/miss ratio • Hacking stats using ftrace (/eBPF): # ./cachestat 1 Counting cache functions... Output every 1 seconds. HITS MISSES DIRTIES RATIO BUFFERS_MB CACHE_MB 210 869 0 19.5% 2 209 444 1413 0 23.9% 8 210 471 1399 0 25.2% 12 211 403 1507 3 21.1% 18 211 967 1853 3 34.3% 24 212 [...]
  • 33. What You Can Do • Verify and understand existing metrics – Even %CPU can be misleading – Cross check with another tool & backend – Test with known workloads – Read the source, including comments – Use "known to be good" metrics to sanity test others • Find missing metrics – Follow the USE Method, and other methodologies – Draw a functional diagram • Burn it all down and start again from scratch?
  • 35. Linux perf • Can sample stack traces and summarize output: # perf report -n -stdio […] # Overhead Samples Command Shared Object Symbol # ........ ............ ....... ................. ............................. # 20.42% 605 bash [kernel.kallsyms] [k] xen_hypercall_xen_version | --- xen_hypercall_xen_version check_events | |--44.13%-- syscall_trace_enter | tracesys | | | |--35.58%-- __GI___libc_fcntl | | | | | |--65.26%-- do_redirection_internal | | | do_redirections | | | execute_builtin_or_function | | | execute_simple_command [… ~13,000 lines truncated …]
  • 37. … as a Flame Graph
  • 39. Java Profilers Java (+object stats) GC Kernel, libraries, JVM CPU Flame Graph
  • 40. Java Profilers • Typical problems: – Sampling at safepoints (skew) – Method tracing observer effect – RUNNING != on-CPU (e.g., epoll) – Missing GC or JVM CPU time • Inaccurate (skewed) and incomplete profiles • Let's try a system profiler?
  • 41. System Profilers with Java (x86) Java (missing stacks & symbols) Kernel TCP/IP GC Idle thread Time Locks epoll JVM compiler optimization #fail
  • 43. Broken System Stack Traces • Broken stacks (1 or 2 levels deep, junk values): • On x86 (x86_64), hotspot reuses the frame pointer register (RBP) as general purpose (a "compiler optimization"), which once upon a time made sense • gcc has -fno-omit-frame-pointer to avoid this – JDK8u60+ now has this as -XX:+PreserveFramePoiner # perf record –F 99 –a –g – sleep 30; perf script […] java 4579 cpu-clock: ffffffff8172adff tracesys ([kernel.kallsyms]) 7f4183bad7ce pthread_cond_timedwait@@GLIBC_2… java 4579 cpu-clock: 7f417908c10b [unknown] (/tmp/perf-4458.map) java 4579 cpu-clock: 7f4179101c97 [unknown] (/tmp/perf-4458.map)
  • 44. • Missing symbols may show up as hex; e.g., Linux perf: • For applications, install debug symbol package • For JIT'd code, Linux perf already looks for an externally provided symbol file: /tmp/perf-PID.map – Find a way to do this for your runtime Missing Symbols # perf script Failed to open /tmp/perf-8131.map, continuing without symbols […] java 8131 cpu-clock: 7fff76f2dce1 [unknown] ([vdso]) 7fd3173f7a93 os::javaTimeMillis() (/usr/lib/jvm… 7fd301861e46 [unknown] (/tmp/perf-8131.map) […]
  • 46. Instruction Profiling # perf annotate -i perf.data.noplooper --stdio Percent | Source code & Disassembly of noplooper ---------------------------------------------------- : Disassembly of section .text: : : 00000000004004ed <main>: 0.00 : 4004ed: push %rbp 0.00 : 4004ee: mov %rsp,%rbp 20.86 : 4004f1: nop 0.00 : 4004f2: nop 0.00 : 4004f3: nop 0.00 : 4004f4: nop 19.84 : 4004f5: nop 0.00 : 4004f6: nop 0.00 : 4004f7: nop 0.00 : 4004f8: nop 18.73 : 4004f9: nop 0.00 : 4004fa: nop 0.00 : 4004fb: nop 0.00 : 4004fc: nop 19.08 : 4004fd: nop 0.00 : 4004fe: nop 0.00 : 4004ff: nop 0.00 : 400500: nop 21.49 : 400501: jmp 4004f1 <main+0x4> • Often broken nowadays due to skid, out-of-order execution, and sampling the resumption instruction • Better with PEBS support
  • 47. What You Can Do • Do stack trace profiling – Get stack traces to work – Get symbols to work – This all may be a lot of work. It's worth it! • Make CPU flame graphs!
  • 49. tcpdump • Packet tracing doesn't scale. Overheads: – CPU cost of per-packet tracing (improved by [e]BPF) • Consider CPU budget per-packet at 10/40/100 GbE – Transfer to user-level (improved by ring buffers) – File system storage (more CPU, and disk I/O) – Possible additional network transfer • Can also drop packets when overloaded • You should only trace send/receive as a last resort – I solve problems by tracing lower frequency TCP events $ tcpdump -i eth0 -w /tmp/out.tcpdump tcpdump: listening on eth0, link-type EN10MB (Ethernet), capture size 65535 bytes ^C7985 packets captured 8996 packets received by filter 1010 packets dropped by kernel
  • 51. strace • Before: • After: • 442x slower. This is worst case. • strace(1) pauses the process twice for each syscall. This is like putting metering lights on your app. – "BUGS: A traced process runs slowly." – strace(1) man page $ dd if=/dev/zero of=/dev/null bs=1 count=500k […] 512000 bytes (512 kB) copied, 0.103851 s, 4.9 MB/s $ strace -eaccept dd if=/dev/zero of=/dev/null bs=1 count=500k […] 512000 bytes (512 kB) copied, 45.9599 s, 11.1 kB/s
  • 53. perf_events • Buffered tracing helps, but you can still trace too much: • Overhead = event instrumentation cost X event frequency • Costs – Higher: event dumps (perf.data), stack traces, copyin/outs – Lower: counters, in-kernel aggregations (ftrace, eBPF) • Frequencies – Higher: instructions, scheduler, malloc/free, Java methods – Lower: process creation & destruction, disk I/O (usually) # perf record -e sched:sched_switch -a -g -- sleep 1 [ perf record: Woken up 3 times to write data ] [ perf record: Captured and wrote 100.212 MB perf.data (486550 samples) ]
  • 55. Valgrind • A suite of tools including an extensive leak detector • To its credit it does warn the end user "Your program will run much slower (eg. 20 to 30 times) than normal" – http://valgrind.org/docs/manual/quick-start.html
  • 57. Java Profilers • Some Java profilers have two modes: – Sampling stacks: eg, at 100 Hertz – Tracing methods: instrumenting and timing every method • Method timing has been described as "highly accurate", despite slowing the target by up to 1000x! • For more about Java profiler issues, see Nitsan Wakart's QCon2015 talk "Profilers are Lying Hobbitses"
  • 58. What You Can Do • Understand how the profiler works – Measure overhead – Know the frequency of instrumented events • Use in-kernel summaries (ftrace, eBPF) – < 10,000 events/sec, probably ok – > 100,000 events/sec, overhead may start to be measurable
  • 60. Monitoring • By now you should recognize these pathologies: – Let's just graph the system metrics! • That's not the problem that needs solving – Let's just trace everything and post process! • Now you have one million problems per second • Monitoring adds additional problems: – Let's have a cloud-wide dashboard update per-second! • From every instance? Packet overheads? – Now we have billions of metrics!
  • 61. STATISTICS "Then there is the man who drowned crossing a stream with an average depth of six inches." – W.I.E. Gates
  • 62. Statistics • Averages can be misleading – Hide latency outliers – Per-minute averages can hide multi-second issues • Percentiles can be misleading – Probability of hitting 99.9th latency may be more than 1/1000 after many dependency requests • Show the distribution: – Summarize: histogram, density plot, frequency trail – Over-time: scatter plot, heat map
  • 63. Average Latency • When the index of central tendency isn't…
  • 65. Traffic Lights RED == bad, GREEN == good …misleading for subjective metrics Better suited for objective metrics
  • 67. Pie Charts …for real-time metrics usr sys wait idle
  • 68. What You Can Do • Monitoring: – Verify metrics, test overhead (same as tools) • Statistics: – Ask how is this calculated? – Study the full distribution • Visualizations: – Use histograms, heat maps, flame graphs
  • 69. BENCHMARKING Benchmarks Common Mistakes Micro Macro Kitchen-Sink bonnie++ Apache Bench
  • 71. ~100% of Benchmarks are Wrong • "Most popular benchmarks are flawed" – Traeger, A., E. Zadok, N. Joukov, and C. Wright. "A Nine Year Study of File System and Storage Benchmarking," ACM Transactions on Storage, 2008. • All alternates can also be flawed
  • 73. Common Mistakes 1. Testing the wrong target – eg, FS cache instead of disk; misconfiguration 2. Choosing the wrong target – eg, disk instead of FS cache … doesn’t resemble real world 3. Invalid results – benchmark software bugs 4. Ignoring errors – error path may be fast! 5. Ignoring variance or perturbations – real workload isn't steady/consistent, which matters 6. Misleading results – Casual benchmarking: you benchmark A, but actually measure B, and conclude you measured C
  • 75. Micro Benchmarks • Test a specific function in isolation. e.g.: – File system maximum cached read ops/sec – Network maximum throughput • Examples of bad microbenchmarks: – gitpid() in a tight loop – speed of /dev/zero and /dev/null • Common problems: – Testing a workload that is not very relevant – Missing other workloads that are relevant
  • 77. Macro Benchmarks • Simulate application user load. e.g.: – Simulated web client transaction • Common problems: – Misplaced trust: believed to be realistic, but misses variance, errors, perturbations, etc. – Complex to debug, verify, and root cause
  • 79. Kitchen Sink Benchmarks • Run everything! – Mostly random benchmarks found on the Internet, where most are are broken or irrelevant – Developers focus on collecting more benchmarks than verifying or fixing the existing ones • Myth that more benchmarks == greater accuracy – No, use active benchmarking (analysis)
  • 81. bonnie++ • "simple tests of hard drive and file system performance" • First metric printed: per character sequential output • What I found it actually tested: – 1 byte writes to libc (via putc()) – 4 Kbyte writes from libc -> FS (depends on OS; see setbuffer()) – 128 Kbyte async writes to disk (depends on storage stack) – Any file system throttles that may be present (eg, ionice) – C++ code, to some extent (bonnie++ 10% slower than Bonnie) • Actual limiter: – Single threaded write_block_putc() and putc() calls • Now thankfully fixed
  • 83. Apache Bench • HTTP web server benchmark • Single thread limited (use wrk for multi-threaded) • Keep-alive option (-k): – without: Can become an unrealistic TCP session benchmark – with: Can become an unrealistic server throughput test • Performance issues of ab's own code
  • 85. UnixBench • The original kitchen-sink micro benchmark from 1984, published in BYTE magazine • Results summarized as "The BYTE Index". Including: • Many problems, starting with… system: dhry2reg Dhrystone 2 using register variables whetstone-double Double-Precision Whetstone syscall System Call Overhead pipe Pipe Throughput context1 Pipe-based Context Switching spawn Process Creation execl Execl Throughput fstime-w File Write 1024 bufsize 2000 maxblocks fstime-r File Read 1024 bufsize 2000 maxblocks fstime File Copy 1024 bufsize 2000 maxblocks fsbuffer-w File Write 256 bufsize 500 maxblocks fsbuffer-r File Read 256 bufsize 500 maxblocks fsbuffer File Copy 256 bufsize 500 maxblocks fsdisk-w File Write 4096 bufsize 8000 maxblocks […]
  • 86. UnixBench Makefile • Default (by ./Run) for Linux. Would you edit it? Then what? • I "fixed" it and "improved" Dhrystone 2 performance by 64% ## Very generic #OPTON = -O ## For Linux 486/Pentium, GCC 2.7.x and 2.8.x #OPTON = -O2 -fomit-frame-pointer -fforce-addr -fforce-mem -ffast-math # -m486 -malign-loops=2 -malign-jumps=2 -malign-functions=2 ## For Linux, GCC previous to 2.7.0 #OPTON = -O2 -fomit-frame-pointer -fforce-addr -fforce-mem -ffast-math -m486 #OPTON = -O2 -fomit-frame-pointer -fforce-addr -fforce-mem -ffast-math # -m386 -malign-loops=1 -malign-jumps=1 -malign-functions=1 ## For Solaris 2, or general-purpose GCC 2.7.x OPTON = -O2 -fomit-frame-pointer -fforce-addr -ffast-math -Wall ## For Digital Unix v4.x, with DEC cc v5.x #OPTON = -O4 #CFLAGS = -DTIME -std1 -verbose -w0
  • 87. UnixBench Documentation "The results will depend not only on your hardware, but on your operating system, libraries, and even compiler." "So you may want to make sure that all your test systems are running the same version of the OS; or at least publish the OS and compuiler versions with your results." … UnixBench was innovative & useful, but it's time has passed
  • 88. What You Can Do • Match the benchmark to your workload • Active Benchmarking 1. Configure the benchmark to run in steady state, 24x7 2. Do root-cause analysis of benchmark performance 3. Answer: why X and not 10X? Limiting factor? It can take 1-2 weeks to debug a single benchmark
  • 90. Observe Everything • Trust nothing. Verify. Write small tests. • Pose Q's first then find the metrics. e.g., functional diagrams: Reference: http://www.brendangregg.com/linuxperf.html
  • 91. • e.g., Java Mixed-Mode Flame Graphs: Profile Everything Java JVM Kernel GC Reference: http://www.brendangregg.com/linuxperf.html
  • 92. Visualize Everything • Full distributions of latency. e.g., heat maps: Reference: http://queue.acm.org/detail.cfm?id=1809426
  • 93. Benchmark Nothing! (if you must, use Active Benchmarking)
  • 94. Links & References • Things that aren't broken: – http://www.brendangregg.com/linuxperf.html • References: – https://upload.wikimedia.org/wikipedia/commons/6/64/Intel_Nehalem_arch.svg – http://www.linuxatemyram.com/ – Traeger, A., E. Zadok, N. Joukov, and C. Wright. “A Nine Year Study of File System and Storage Benchmarking,” ACM Trans- actions on Storage, 2008. – http://www.brendangregg.com/blog/2014-06-09/java-cpu-sampling-using-hprof.html – http://www.brendangregg.com/activebenchmarking.html – https://blogs.oracle.com/roch/entry/decoding_bonnie – http://www.brendangregg.com/blog/2014-05-02/compilers-love-messing-with- benchmarks.html – https://code.google.com/p/byte-unixbench/ – https://qconsf.com/sf2015/presentation/how-not-measure-latency – https://qconsf.com/system/files/presentation-slides/profilers_are_lying_hobbitses.pdf – Caution signs drawn be me, inspired by real-world signs
  • 95. Thanks • Questions? • http://techblog.netflix.com • http://slideshare.net/brendangregg • http://www.brendangregg.com • [email protected] • @brendangregg Jan 2016

Editor's Notes

  • #2: G'Day, I'm Brendan. I normally give talks about performance tools that work, but today I'm going to talk about those that don't: broken and misleading tools and metrics.
  • #4: Learn that everything can be broken or misleading. Popularity is not credibility.
  • #5: Awesome place to work Centos and Ubuntu
  • #6: Useful exercise for extracting new metrics from the system. Complementary to observability. Since benchmarking can perturb, can be tried after observability. Accurate and realistic benchmarking is vital for industry investments. However, benchmarking is error prone
  • #11: Is this ok? Does that answer our questions? Pretty simple, right?
  • #17: A CPU mystery: As load increased, CPU ms per request lowered (blue) up to 1.84x faster Was it due to: Cache warmth? no Different code? no Turbo boost? no
  • #34: Cross check Linux ftrace, my perf-tools, are good for ad-hoc checks. funccount, kprobe, … Test with known workloads Do the numbers add up? Try a prime number, eg, 23. Read the source The source may say "/* XXX: broken when feature X is turned on. Fix. */" Starting again from scrath unikernels
  • #48: Stack traces -fno-omit-frame-pointer, -XX:+PreserveFramePointer If needed, get PEBsSto work
  • #59: Even if the overhead is high, it may be worth it, provided it is known beforehand
  • #66: imagine you went to another country and hired a car, and were told that "green" on a traffic light means "probably ok, but you might get T-boned"… That'd make the traffic lights pretty useless.
  • #70: Useful exercise for extracting new metrics from the system. Complementary to observability. Since benchmarking can perturb, can be tried after observability. Accurate and realistic benchmarking is vital for industry investments. However, benchmarking is error prone
  • #87: I saw Solaris 8 in the SCaLE Computer Museum at Game Night last night
  • #94: If you must, trust nothing, verify everything, and use Active Benchmarking
  • #96: We're hiring