Re: About the web polygraph to test squid performance

Hi alex:

Thank you for your help. Today I setup the polymix4 to test. But I meet
the weird error :

./../src/runtime/IOBuf.h:37: assertion failed: 'theInOff <= theCapacity'

in server side. The whole log and configuration file is enclosed.

And I still have several question:

1. How can I tell whether how many paires client/server can give enough
pressure to my squid?

2. how long the test will last and how can I control it ?

Best Regards

Leon Li *RHCE* 708 Tower A, Fortune Plaza, No. 7
Technical Account Manager Dongsanhuan Zhonglu 100020
Red Hat,Greater China Phone: +86-10-65339318
Email: xli_at_redhat.com Fax +86 10 6533 9400

Alex Rousskov wrote:
> On Fri, 2007-06-01 at 13:14 +0800, Leon Xiang Li wrote:
>
>> 1. I should set the client_side.hosts to my polyclt local network
>> address ( such as eth0's IP ) and server_side.hosts to my polysrv's
>> local network address , right ?
>
> Yes.
>
>> and how polygraph use the info?
>
> Please see the following document (especially Section 2.2) if you want
> to know the details:
> http://www.web-polygraph.org/docs/userman/address.html
>
>> and in howto, it read a IP range, why ?
>
> Many tests involve more than one client-server pair, so you may have
> multiple client and server host addresses. The official PolyMix-4 limit
> is 500 requests/second per pair. If your cache can handle more than
> that, you need more than one pair or you need to make a custom workload
> with a higher limit.
>
>> 2. In howto, it read that the polygraph will add many ip alias to my lo
>> (lo::10.101.0-123.1-250/22)
>>
>> but in order to use it. I need add the proper route rule to my linux box
>> which show how to do it in bsd.
>>
>> I am not sure how to do in linux. And should I add the route rule on all
>> the three servers?
>
> You should add appropriate routes on all hosts. AFAIK, Linux route
> command has very similar but different syntax compared to FreeBSD route.
> IIRC, you need to prefix the gateway address with 'gw'.
>
> As a rule of thumb, if ping from {robot, proxy, server} IP does not
> reach {robot, proxy, server} IP, you are not done.
>
> HTH,
>
> Alex.
>
>
>> Alex Rousskov wrote:
>>> On Fri, 2007-06-01 at 09:05 +0800, Leon Xiang Li wrote:
>>>> Thanks a lot for your helpful reply. I will try mix-4 instead .
>>>>
>>>> Actually I just set a simple test: three server , one running squid ,
>>>> another two running ploysrv , poylclt seperately.
>>>>
>>>> Can you give me some clue about how to modify PolyMix-4.pg?
>>>>
>>>>
>>>> how can I set IP ?
>>>>
>>>> client_side = {
>>>> addr_space = [ 'lo0::10.101.0-123.1-250/22' ];
>>>> hosts = [ '172.16.101.61-62' ]; // replace with IPs of
>>>> your PCs
>>>> };
>>>> server_side = {
>>>> addr_space = [ 'lo0::10.101.128-251.1-250:80/22' ];
>>>> hosts = [ '172.16.101.191-192' ]; // replace with IPs of
>>>> your PCs
>>>> };
>>>>
>>> You can usually leave the addr_space intact (unless the hard coded
>>> addresses conflict with other local addresses or the interface name is
>>> wrong). The hosts need to be set to the primary IP addresses of the
>>> client and server box. The HOWTO describes these modifications in
>>> Section 5:
>>>
>>> http://www.measurement-factory.com/docs/PolyMix-4/
>>>
>>>> What is the reasonable value for peak_req_rate & FillRate ?
>>> Depends on your Squid and its box. I would start with 10 requests per
>>> second, just to make sure everything works. Filling at that rate will
>>> take for ever so stop the test once you see that it is making progress
>>> without errors. See cache-off results for some ideas regarding Squid
>>> (and other proxy) performance; adjust to take into the account modern
>>> hardware:
>>>
>>> http://www.measurement-factory.com/results/
>>>
>>> As for the fill rate, you can try setting it to 100% of peak, but Squid
>>> may not survive mostly-misses loads if your peak rate is high. 50% is
>>> probably safe; you may want to double check with Squid configs at the
>>> cache-off. Please note that the value may depend on your Squid cache_dir
>>> type.
>>>
>>>> How about if I have no dns server in my private network?
>>>> Can I ignore it?
>>> You can disable DNS. Searching Polygraph site yields the following email
>>> with specific instructions. I hope it is still valid:
>>> http://www.web-polygraph.org/mail-archive/users/200203/0004.html
>>>
>>> Good luck,
>>>
>>> Alex.
>>>
>>>
>>>> Alex Rousskov wrote:
>>>>> On Thu, 2007-05-31 at 20:19 +0800, Leon Xiang Li wrote:
>>>>>
>>>>>> I use polymix-1.pg to test the squid preformance.
>>>>> You really should not. Please use PolyMix-4 or, at the very least,
>>>>> PolyMix-3.
>>>>>
>>>>>> But there is no much document available about the meaning of pgl
>>>>>> configuration file and working theory of polygraph.
>>>>> You will find workload descriptions and PGL reference at
>>>>> http://www.web-polygraph.org/docs/workloads/
>>>>> http://www.web-polygraph.org/docs/reference/pgl/
>>>>>
>>>>> The HOWTO documents and cache-off reports linked from the workload pages
>>>>> may be very useful as well. You will find all our public benchmarking
>>>>> results at http://www.measurement-factory.com/results/
>>>>>
>>>>> Finally, Appendix D of Duane Wessel's Squid book talks about Squid
>>>>> performance testing with Polygraph: http://squidbook.org/
>>>>>
>>>>>> But after 1 hour, it still runing and in i-warm phase.
>>>>> I do not remember much about PolyMix-1 (it was written even before PGL),
>>>>> but it is possible that your robots cannot contact all the origin
>>>>> servers and keep trying. If you insist on using PolyMix-1, please use
>>>>> verbosity level 10 and post console logs for further assistance. I would
>>>>> not recommend relying on PolyMix-1 results though.
>>>>>
>>>>> Thank you,
>>>>>
>>>>> Alex.
>>>>>
>>>>>
>
Received on Mon Jun 04 2007 - 08:24:34 MDT