Scribd
Upload
120 views21 pages

Varnish: - The HTTP Accelerator Poul-Henning Kamp

This document provides an overview of Varnish, an HTTP accelerator for caching dynamic web content. It describes Varnish as being focused specifically on improving server-side speed for content management systems. The key features highlighted include high performance, content provider features like instant URL invalidation, and a configuration language called VCL. Logging and statistics collection is done via shared memory for low overhead. Varnish is open source and was originally developed and funded by a Norwegian newspaper.

Uploaded by

mdemark
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
120 views21 pages

Varnish: - The HTTP Accelerator Poul-Henning Kamp

This document provides an overview of Varnish, an HTTP accelerator for caching dynamic web content. It describes Varnish as being focused specifically on improving server-side speed for content management systems. The key features highlighted include high performance, content provider features like instant URL invalidation, and a configuration language called VCL. Logging and statistics collection is done via shared memory for low overhead. Varnish is open source and was originally developed and funded by a Norwegian newspaper.

Uploaded by

mdemark
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 21

Varnish

- the http accelerator

Poul-Henning Kamp

[email protected]
CMS systems
● GET / HTTP/1.1
– Hang on, I need to look up a few hundred
things in my database and then do a lot of
editing with some badly written software.
CMS systems
● GET / HTTP/1.1
– Hang on, I need to look up a few hundred
things in my database and then do a lot of
editing with some badly written software.

● HEAD / HTTP/1.1
– Hang on, I need to look up a few hundred
things in my database and then do a lot of
editing with some badly written software, and
then I will throw the result away.
CMS systems are SLOW!
● Complex content generation process
● Single database prevents clustering
● Expensive software ditto.

● => Need for server side caching.


– Apache
– Squid
– $CALL
Apache as cache
● Not what Apache is built for.
● Not what Apache is good at.
● Square peg, round hole.
Squid is a cache...
● ...built for client side caching.
● Lots of unwanted functionality.
– Authentication.
– FTP
● 1980 software design.
● Fragile and buggy.
● Wrong configuration features.
● Wrong policy decisions.
$CALL
● Akamai will happily take your money.
● But you loose control.
● You're stuck with their service.
dictionary:Varnish
● tr. v. var·nished, var·nish·ing, var·nish·es

– 1. To cover with varnish.

– 2. To give a smooth and glossy finish to.

– 3. To give a deceptively attractive


appearance to.
Varnish Cheat-Sheet
● Web-accelleration for slow CMS systems
● Narrow focus on server side speedup
– No FTP etc.
– Content provider features
● High Performance
– 32 & 64bit, large RAM, sendfile, accept filters
– SMP/Multicore friendly architecture
– 2006 software design
– 11 syscall + 7 locks per cache hit
Performance Design
● Don't copy data if we can avoid it
– Avoid text-processing headers
● Convert Chunked encoding to Direct
● Also cache ”cannot be cached” info
● Maximize session usage
– Pass-through mode understands chunked
encoding etc.
– Pipe mode for weird stuff (selected by VCL)
Varnish is fast!
● 1000 cache hits @ 100 Mbit/sec:
Sample Min Max Median Average Stddev
Full 12,9 3001 16,2 71,1 338,5
90% fractile 12,8 26 15,9 16,3 1,7
(all times are in microseconds)
Content Provider Features
● Instant URL invalidation
– Regexp matching
– Easy Integration to CMS system
● Expiry time control (via VCL)
● Load/Situation mitigation (via VCL)
OaM
● Cluster controller (version 2 feature)
– Invalidate URL on all caches instantly
– Change VCL on all caches simultaneously.
– Aggregate statistics

● Command Line Interface


– Remote access via ssh/mobile phone
– Interrogation/configuation/investigation
Varnish Config Language
● Simple domain specific language
– Compiled via C language to binary
● Transparantly!
– Dynamically loaded
● Multiple configs loaded concurrently
– Instant switch from one VCL to another.
VCL example
if (client.ip in 10.0.0.0/8) {
pass;
}

if (req.url.host ~ "cnn.no$") {
rewrite req.url.host ”cnn.no” ”vg.no”
}

if (!backend.up) {
if (obj.exist) {
set obj.ttl += 10m;
deliver;
}
switch_config ”ohhshit”;
}
Logging
● Logging to shared memory
– No slowdown for real workload

● Daemons tailing shm generate ”real”


output:
– Apache format
– Custom format
– Realtime views
”Varnishtop”
● Shows an real-time view of log records.
● ”varnishtop -i rxurl”
342.49 /gfk/front/tipsvg.png
340.41 /sistenytt.html
326.53 /gfk/ann/n.gif
307.61 /annonser/telenor/menyknapp/svart/vg115x160_1.html
297.70 /include/global/art.js
276.73 /gfk/ann/ng.gif
273.84 /css/hoved.css
243.01 /
241.85 /annonser/telenor/menyknapp/svart/pil.gif
226.04 /css/kjaerevg.css
”varnishhist”
● Shows response time as real-time
histogram:
    
   |
   |
  ||
  ||
  ||
  ||
  ||
  ||
||||||||||||||||||||| ||#||||##   # ##  #  # #          # #
+­­­­­­­­­+­­­­­­­­­+­­­­­­­­­+­­­­­­­­­+­­­­­­­­­+­­­­­­­­­+­­­­­­­­­+­
|1e­5     |1e­4     |1e­3     |1e­2     |1e­1     |1e0      |1e1      |1
Statistics
● Stored in shared memory

● Multiple programs can monitor, react to


and & present stats data.
– At practically no cost in cpu time.
Varnish 1.0
● Sponsored by Verdens Gang
– Norvegian Newspaper
● 500.000 Kr ($85K)
● 22000 lines of code
– 22 Kr/line ($3.80/line)
● 1000 checkins
– 500 Kr/checkin ($85/checkin)
Varnish 1.0 Cheatsheet
● WWW.varnish-cache.org
– Yes, we need to work on the web-page.
● Coded by: Poul-Henning Kamp
– [email protected]
● Runs on: Any resonable modern UNIX
– FreeBSD, Linux etc.
● Open Source
– BSD license

You might also like