The ilities of software engineering.pptx
0 likes43 views
The document discusses the importance of non-functional requirements, referred to as '-ilities', in driving innovations in managed runtimes and software engineering. It emphasizes aspects such as performance, scalability, usability, and portability, detailing how these qualities influence system architecture and design decisions. The content also explores specific scenarios related to garbage collection and system scalability, providing insights into performance criteria and the concurrent handling of tasks in a distributed environment.
The ilities of software engineering.pptx
- 1. HOW QUALITIES THAT GO BEYOND FUNCTIONAL REQUIREMENTS DRIVE INNOVATIONS IN MANAGED RUNTIMES THE "-ILITIES" OF SOFTWARE ENGINEERING Monica @mon_bec k
- 3. THE AGENDA Discuss “-ilities” aka non- functional requirements and dive into Performance with GC architecture + Scale the Clouds and provide some awesome quotes
- 4. Software Engineering What are we building? How are we building it?
- 6. Software Engineering Functional Requirements Qualities aka “ilities” aka Non-Functional Requirements
- 7. “ILITIES”? Ility is an abstract quality that good software should exhibit - Wiktionary
- 11. I WANT EVERYTHING TO GO FAST(ER)! PERFORMAN CE
- 12. WHAT DOES THAT EVEN MEAN?
- 15. A pplication timeline T0 T1 Stim u l u s arriv al Respo nse sent
- 17. A pplication tim eline Stim ulus arriv al Respo nse sent T0 T1 A pplication timeline T0 T2 T1 Stim u l u s arriv a l Respo nse sent
- 18. A pplication timeline T0 T1 A pplication tim eline T0 T2 T1
- 19. Performance How?
- 21. THROUGHPUT AND LATENCY ARE THE TWO MAIN DRIVERS TOWARDS REFINEMENT IN GARBAGE COLLECTION ALGORITHMS GC FACT:
- 25. Parallel GC Z GC or Shenando ah GC OPENJDK GCS
- 27. Ility How? What?
- 29. SHOULD I SCALE UP OR SHOULD I SCALE OUT? SCALABILITY
- 30. Pros Cons Bigger system – more processing power, higher bandwidth, spatial locality Hardware layout and data temporal locality needs could lead to latency increase and non- uniformity of access Almost little to none configuration changes lead to faster turnaround time The workload’s bulk and scale could render it to work only with ageing systems Scaling Up in System Pros Cons Highly available system - downtime is non-existent Study your system’s scaling factor Grow or shrink the distributed footprint on demand Higher footprint would add more costs (power, space, cooling) Scaling Out with System
- 31. Scalability How?
- 32. SCALING THE CLOUD FOR PERFORMANCE
- 33. Core Counts Scaling Factor Max System Capacity - Demand driven - Highlights concurrency effects - Helps with Scaling Up or Scaling Out decisions - Iterative process - Determines max throughput - Helps with understanding the number of concurrent users or requests that a system can handle Incrementing the worker threads
- 34. 0 0.5 1 1.5 2 2.5 -0.2 0 0.2 0.4 0.6 0.8 1 1.2 2 4 8 16 24 32 48 64 96 Normalized Gen 2 Max System Capacity Gen 2 Scaling Factor Core Counts Scaling Factor Max System Capacity Core Counts
- 35. 0 0.5 1 1.5 2 2.5 -0.2 0 0.2 0.4 0.6 0.8 1 1.2 2 4 8 16 24 32 48 64 96 Normalized Gen3 Max System Capacity Gen 3 Scaling Factor Scaling Factor Max System Capacity Core Counts
- 36. 0 0.2 0.4 0.6 0.8 1 1.2 Single JVM Dual JVM Quad JVM Normalized Gen 3 Max System Capacity Max System Capacity
- 38. -0.2 0 0.2 0.4 0.6 0.8 1 1.2 2 4 8 16 24 32 48 64 96 Non-Linear Non-Linear Trendline
- 39. -0.2 0 0.2 0.4 0.6 0.8 1 1.2 2 4 8 16 24 32 48 64 96 Non-Linear Linear Non-Linear Trendline Linear Trendline
- 40. -0.5 0 0.5 1 1.5 2 2.5 3 2 4 8 16 24 32 48 64 96 Non-Linear Linear Non-Linear Trendline Linear Trendline
- 41. -1 0 1 2 3 4 5 2 4 8 16 24 32 48 64 96 Non-Linear Linear Non-Linear Trendline Linear Trendline
- 42. -1 0 1 2 3 4 5 2 4 8 16 24 32 48 64 96 Non-Linear Linear These projections look crazy, don’t they?
- 46. A FEW INNOVATIONS IN OPENJDK DRIVEN BY NON - FUNCTIONAL REQUIREMENTS
- 47. Portability What? Write Once Run Anywhere... ? How?
- 49. WHAT IS AN OPENJDK PORT? • Windows on Arm64 was a new platform for OpenJDK • In-order to run Java applications on this new platform, we need to get the Java Runtime Environment (JRE) to work on it. • A JDK is a superset of a JRE as it includes tools and utilities for development and debugging • In order to be able to develop in Java, we need to get the JDK to work on the new platform Java HotSpot VM A JRE Class Libraries Runtime Execution Engine UI Toolkits Base, Lang, Util Libs
- 50. Usability What?
- 52. Customer Experience = Capability + Reliability + Security - Alan Chan Managing Director, Global Head of SRE, JP Morgan Chase & Co
- 55. DO MORE WITH LESS - John Cunningham CVP, Software Engineering, Microsoft
- 57. THANK YOU
- 58. Microsoft’s commitment to the Java ecosystem
- 60. LinkedIn Azure Yammer Minecraft Realms Bing * Internal systems. Does not include customer workloads. Others … Java is widely used internally at Microsoft
- 62. CONTENTION
- 63. THROTTLING
- 64. SCALABILITY
- 65. LET’S TALK ABOUT CONCURRENCY & PARALLELISM This Photo by Unknown Author is licensed under CC BY-NC-ND
- 66. Concurrency = Throughput x Response Time
- 67. JAVA MUTATORS AND GC THREADS WORKING TOGETHER
- 68. G1 GC Parallel Full GC
- 69. GC Background Threads Java Application Threads Paus e Mark Start Paus e Mark End Pause Relocat e Start Concurrent Mark/Rema p Concurrent Relocate Concurrent Prepare for Relocation GC Background Threads CONCURRENCY & PARALLELISM IN Z GC
Editor's Notes
- #3: Jakub’s talk had cat photos and it was awesome and brought a smile to my face, so here I have my cats and dogs. I hope they bring smile to yours as well
- #5: So let’s start from the very beginning – Functionality – what it does Operation – how it does it – The ability of a software to execute and evolve
- #6: Functionality – what it does Operation – how it does it – The ability of a software to execute and evolve
- #7: Functionality – what it does Operation – how it does it – The ability of a software to execute and evolve
- #8: Say you’d like to buy a car or a truck – and you’d like to be able to tow a trailer – now that becomes a desired attribute for useability and enables scalability
- #9: In this talk, I’d like to walk a few of this attributes with you Security – e.g. Transport Layer Security (TLS 1.3 vs 1.0) Observability: ability to understand complex systems internal state based on workloads, e.g. JFR
- #10: and together we can investigate the What the How. First we will dive into designing a GC for performance, then we will look at Scalability and investigate VMs for either scale out or scale up purposes
- #11: let’s dive in deeper to performance requirements by trying to define the attributes first
- #12: so what’s the first thing that people talk about when discussing performance? Say if you ask your end user “what needs to happen inorder for our system to be performant?” what would their answer look like?
- #13: Let’s get some clarity on that
- #14: But make sure you define what fast means to your customer. Refine definition This is gathering details wrt nfr.
- #15: Footprint is the space that is needed to run your software.It's both about the memory footprint and also the physical resources Responsiveness: If I send a stimulus now, how much time ’til I get a response back? Throughput: How can I maximize the operations per second of my system? Availability is the uptime and Mean Time Between Failure of the system Let’s start with what we mean by response time.
- #16: If I send a stimulus now, how much time ’til I get a response back?
- #18: How can I maximize the operations per second of my system?
- #19: Work done is Throughput = operations per second. (transactional DB) transactions per second, (query DB): requests per second
- #20: Now that we know some of the attributes, let’s try to understand the how
- #21: Before we start with our GC architecture, let me give you a fact that will help clarify why I chose GC for the perf requirement exercise.
- #23: So, what do we know – we know that the GC will need to do some GC work and then we also know the weak generational hypothesis that most objects will die young.
- #24: No concurrency == app has max utilization of resources and that should give it good throughput
- #25: Regions == smaller work units. More granularity. All concurrent work needs maintenance barriers. With that we have successfully architected two GCs
- #28: Next ility
- #30: What’s the age-old question that everyone wants to know?
- #32: Now that we know some of the attributes, let’s try to understand what they stand for and then let’s dive in deeper to performance requirements by investigating the how
- #34: Normalized =(C18-MIN(C18:C25))/(MAX(C18:C25)-MIN(C18:C25))
- #35: At 1, it looks like it could handle the number of concurrent users or requests to fit in with the other cores At 2 and at 3, we see concurrency effects Normalized =(C18-MIN(C18:C25))/(MAX(C18:C25)-MIN(C18:C25))
- #37: Single JVM for gen 3 performs just as well as dual JVM on Gen 2 – so basically Gen 2 will benefit from scale out and Gen 3 will benefit from scaling up to the max cores.
- #44: Gene Amdahl shows that there is a fraction of the workload which cannot be parallelized, called serial fraction i.e. no matter how many cores are allocated for the program, only a single core will be executing the serial portion of the program. We use σ to denote this serial fraction of workload. Hence (1-σ) becomes the parallel fraction of workload. σ takes values in the range (0, 1). Continuous validity of single processor
- #45: Universal Scalability Law (USL) is an extension of Amdahl’s law. It accounts for the additional overhead due to interprocess communication that happens application software level, middleware level, operating system level, and hardware level. Three Cs: Concurrency (gamma), Contention (alpha), Coherency (beta)
- #46: Universal Scalability Law (USL) is an extension of Amdahl’s law. It accounts for the additional overhead due to interprocess communication. Interprocess communication happens at multiple levels within the system: application software level, middleware level, operating system level, and hardware level.
- #48: J/ava is portable. What does it mean for openjdk?
- #50: Executables (In the bin/ subdirectory) An implementation of the Java Runtime Environment (JRE). The JRE includes a Java Virtual Machine (JVM™), class libraries, and other files that support the execution of programs written in the Java programming language. This directory also includes tools and utilities that will help you develop, execute, debug, and document programs written in the Java programming language. For further information, see the tools documentation at https://docs.oracle.com/en/java/javase/12/tools/tools-and-command-reference.html Configuration files (In the conf/ subdirectory) Files that contain user-configurable options. Files in this directory can be edited to change the JDK's access permissions, configure security algorithms, and set the Java Cryptography Extension Policy Files which might be used to limit the JDK's cryptographic strength. C header Files (In the include/ subdirectory) C-language header files that support native-code programming with the Java Native Interface and the Java Virtual Machine (JVM) Debugger Interface. Compiled Java Modules (in the jmods/ subdirectory) Compiled modules used by jlink to create custom runtimes. Copyright and License files (in the legal/ subdirectory) License and copyright files for each module. Includes third party notices as .md (markdown)files. Additional Libraries (In the lib/ subdirectory) Additional class libraries and support files required by the JDK. These files are not intended for external use.
- #51: In the opening keynote we heard about project Amber and how removing the boiler plate is making the Java language much more usable. So let’s briefly look at Usability
- #52: Usability is user-friendliness be it about language level boiler plate removal or having GC with minimal to none command line options or if your error messages like nullpointerexceptions in JDK14 User Experience is perception and reaction to a feature – like some people may prefer Java 19 over Kotlin due to the “User Experience” Customer experience is about the total journey – you will have various touchpoints with your customer. I have this awesome quote by Alan who is the head of SRE @JP Morgan https://www.testingtime.com/en/blog/usability-ux-and-cx-compared/
- #53: 'nfr' needs more love and appreciation from engineers and products. at the end of the day "customer experience = capability + reliability + security" - all are equally critical. a lot of manpower and brainpower are drained today due to support work and manual toils. to make a platform more reliable, those time/resources/power will be reinvested to innovation and creation...and better products.
- #54: As we say our goodbyes today, I’d like to leave you with a few thoughts -
- #55: Throughout this conference and this talk there was a theme of efficiency. So I’d like to say that Ilities are Qualities that make us efficient. So, I hope you internalize them.
- #56: They are asking you to be efficient. Measure in all dimensions Efficiency is defined as the ratio between speed up and the number of processors E(p) = S(p)/P
- #57: Concurrency is one of those requirements that might as well be specified with f
- #61: Microsoft ported OpenJDK to Windows on Arm to support Azure servers, Minecraft, and other scenarios. Minecraft coming soon to Windows on Arm devices. LinkedIn hundreds of thousands of JVMs powering every feature of the service. Bing: indexing services powered by OpenJDK. Bing is the search engine on Windows, behind “Search results” in the Start menu. Minecraft Realms: Microsoft-hosted servers of Minecraft for world-wide community of Minecraft players. Yammer: creators of Java frameworks like Dropwizard. Powers certain features behind Office 365/MS Teams. Azure: infrastructure control plane leverages some Java-based services for resource provisioning orchestration. Nuance and Metaswitch: Recent Microsoft acquisitions heavily invested in Java technologies
- #62: Let’s get some clarity on that
- #63: Concurrency is one of those requirements that might as well be specified with f
- #64: Concurrency is one of those requirements that might as well be specified with f
- #65: Concurrency is one of those requirements that might as well be specified with f
- #66: Concurrency is one of those requirements that might as well be specified with
- #68: Concurrency is one of those requirements that might as well be specified with f