Re: 答复: [pLog-svn] Stress testing tools
Oscar Renalias
oscar at renalias.net
Tue May 16 21:50:10 GMT 2006
Before I managed to kill devel.lifetype.net with siege (by mistake, I
forgot to escape a URL that contained the '&' character and ended up
spawning too many 'siege' processes... whoops), I found some
interesting results:
== LT 1.1, visiting one permalink only ==
Command: siege http://devel.lifetype.net/index.php?
op=ViewArticle&blogId=19&articleId=125
Template cache: On
Data cache: On
Transactions: 3054 hits
Availability: 99.12 %
Elapsed time: 840.70 secs
Data transferred: 57.87 MB
Response time: 3.79 secs
Transaction rate: 3.63 trans/sec
Throughput: 0.07 MB/sec
Concurrency: 13.77
Successful transactions: 3054
Failed transactions: 27
Longest transaction: 30.00
Shortest transaction: 0.62
This test was some sort of 'slashdot' effect, 15 clients making
requests to one single URL as fast as possible from my home
connection. We got 3.63 (good) requests per second with a response
time of 3.79 (not too good) This means about 227 requests per minute,
while an slashdot effect can vary between 100 and 300 (according to
some articles that I found about it)
These figures are not too bad, the only problem being is that loading
the permalink above is not just one hit but a lot more... There are
icons, stylesheets and other stuff that needs to be loaded so that
would be substracted from the transaction rate.
The server load peaked at something like 19.5 or so.
== LT 1.1, summary.php only ==
Command: siege http://devel.lifetype.net/summary.php
Template cache: On
Data cache: On
Transactions: 1741 hits
Availability: 100.00 %
Elapsed time: 258.55 secs
Data transferred: 51.61 MB
Response time: 1.72 secs
Transaction rate: 6.73 trans/sec
Throughput: 0.20 MB/sec
Concurrency: 11.59
Successful transactions: 1741
Failed transactions: 0
Longest transaction: 11.86
Shortest transaction: 0.65
For whatever reason summary.php had a bigger throughput and
therefore, according to siege, the transaction rate was a lot higher
and with a lower response time than before. I don't really know how
to read this data, but we should figure out why summary.php is able
to respond a lot faster to requests than index.php...
As soon as the dev server is up I will setup a 1.0.4 instance with
the same data and see whether these figures are better, the same or
worse. There are also a lot more tests that can be run and hopefully
I will be able to prepare a nice performance report soon.
Oscar
On 16 May 2006, at 20:07, subaochen wrote:
>
> maybe you want to try siege?
>
>> -----邮件原件-----
>> 发件人: plog-svn-bounces at devel.lifetype.net
>> [mailto:plog-svn-bounces at devel.lifetype.net] 代表 Christoph
>> Feddersen
>> 发送时间: 2006年5月16日 0:49
>> 收件人: plog-svn at devel.lifetype.net
>> 主题: Re: [pLog-svn] Stress testing tools
>>
>>
>> I use JMeter most of the time. Easy to use and very flexible.
>>
>> http://jakarta.apache.org/jmeter/
>>
>> Oscar Renalias wrote:
>>> Does anybody have any experience with [free] tools to
>> stress-test web
>>> applications? Apachebench is ok, but is there anything else
>> out there?
>>> Is any of you familiar with any of these?
>>>
>>> http://opensourcetesting.org/performance.php
>>>
>>> We would need something that can play several "scripts"
>> simultaneously.
>>> I doesn't really make any sense to send 1000 requests to
>> summary.php
>>> with caching turned on because the page is generated once
>> and served
>>> cache after that, and this doesn't really represent a normal usage
>>> pattern in a normal blogging community.
>>>
>>> Oscar
>>> _______________________________________________
>>> pLog-svn mailing list
>>> pLog-svn at devel.lifetype.net
>>> http://devel.lifetype.net/mailman/listinfo/plog-svn
>> _______________________________________________
>> pLog-svn mailing list
>> pLog-svn at devel.lifetype.net
>> http://devel.lifetype.net/mailman/listinfo/plog-svn
>>
>
>
>
> _______________________________________________
> pLog-svn mailing list
> pLog-svn at devel.lifetype.net
> http://devel.lifetype.net/mailman/listinfo/plog-svn
>
More information about the pLog-svn
mailing list