Skip to content

Latest commit

 

History

History
103 lines (71 loc) · 3.34 KB

README.md

File metadata and controls

103 lines (71 loc) · 3.34 KB
Raw

Introduction

I wrote that code because: (the obvious reason::I love to write code in Go)

We are working so hard to optimize our servers - why shouldn't we do it for our clients testers?!

I noticed that the existing tools for benchmarking/load HTTP/HTTPS servers has some issues:

  • ab (ApacheBenchmark) - maximum concurrency is 20k (can be eliminate by opening multiple ab processes)
  • Siege are work in a model of native thread per request, meaning that you cannot simulated thousands/ten of thousands clients concurrently even if you tweak the RLIMIT of stack usage per native thread - still that kill the client machine and cause it to be very load and not an efficient client/s. What we really want is minimum resources usage and get the maximum throughput/load!

If you already familiar with the model of Go for high performance I/O with goroutines, we can achieve that mission easily.

The funny part - I did some benchmark to the client tester tool and not to the server:

##Siege vs GoBench:

###Siege:

$>siege -b -t10S -c500 http://localhost:80/

** SIEGE 2.70
** Preparing 500 concurrent users for battle.
The server is now under siege...
Lifting the server siege...      done.
Transactions:		       74247 hits
Availability:		      100.00 %
Elapsed time:		        9.62 secs
Data transferred:	       96.58 MB
Response time:		        0.06 secs
Transaction rate:	     7717.98 trans/sec
Throughput:		       10.04 MB/sec
Concurrency:		      490.19
Successful transactions:       74247
Failed transactions:	           0
Longest transaction:	        1.02
Shortest transaction:	        0.00

###GoBench:

$>gobench -k=true -u http://localhost:80 -c 500 -t 10
Dispatching 500 clients
Waiting for results...

Requests:                           343669 hits
Successful requests:                343669 hits
Network failed:                          0 hits
Bad requests failed (!2xx):              0 hits
Successfull requests rate:           34366 hits/sec
Read throughput:                  54700061 bytes/sec
Write throughput:                  4128684 bytes/sec
Test time:                              10 sec
  • requests hits and requests rate are 5X better on the same time (10 seconds) and the same number of clients (500)!
  • I tried the same with 2000 clients on Siege with proper system configuration, and Siege was crashed
  • I tried gobench with the maximum number of clients that we can use (65535 ports) - it's rocked!
  • Yet I didn't put the results of ab because I still need to investigate the results

Usage

  1. install Go env follow: https://golang.org/dl/

  2. download gobench

    GOPATH=/tmp/ go get github.com/valyala/fasthttp
GOPATH=/tmp/ go get github.com/cmpxchg16/gobench

  3. run some http server on port 80

  4. run gobench for HTTP GET

    $>gobench -u http://localhost:80 -k=true -c 500 -t 10

  5. run gobench for HTTP POST

    $>gobench -u http://localhost:80 -k=true -c 500 -t 10 -d /tmp/post

Notes

  1. Because it's a test tool, in HTTPS the ceritificate verification is insecure
  2. use Go >= 1.5 (fasthttp require it)

Help

gobench --help

License

Licensed under the New BSD License.

Author

Uri Shamay ([email protected])