[httperf] Re: Connection rate is very low

david at lang.hm david at lang.hm
Mon Oct 22 14:14:22 PDT 2012


I do a lot of work through proxies of various types (but do not have 
experience with haproxy), and a very large number of the proxies are 
actually significantly slower than the servers that they are supporting 
nowdays.

Having a proxy cross into overload and have it's performance drop 
significantly as you ramp up traffic is not uncommon.

David Lang


  On Mon, 22 Oct 2012, Vikash Kumar wrote:

>    Thanks for reply. I had already tested this. With demand rate of 1000
> conn/sec we are getting 1000 conn/sec. But with increase increasing
> connection rate to 2000 conn/sec , the result drops  to 500 conn/sec, 482
> req/sec and 482 reply/sec.
>
>    For a while, let us assume that because we are using Haproxy that's why
> we are getting the performance figure very low.
>
>   But logically, the connection rate should not drop this much. Their is
> no comparison between 27K reply rate and 500 reply rate.
>
>  ** It has been proved that haproxy is able to sustain up to 50K
> connections.
>
>   So I am unable understand where the things are getting wrong?
>
> Thanks
>
>
>
> On Mon, Oct 22, 2012 at 6:01 PM, Tim Terrill <TTerrill at synacor.com> wrote:
>
>>   Based on your tests without the proxy responding at 27,000 requests, I
>> think its pretty clear that there's a bottleneck in the configuration that
>> has the proxy.
>>
>>  Have you tried lower rates?  (side note, at that rate, you could only
>> sustain it for 3-4 seconds before running out of TCP Ports (the 65535 max
>> number).  Eventhough you have FD set high, there's only a small limited
>> number of ports available for transmitting.)
>>
>>  I'd say try a rate of 1000, then ramp it up.
>>
>>  To help with that, I'd recommend getting auto bench (google it) which is
>> a perl wrapper that allows you to define start rate, stop rate, step rate
>> so you can slowly ramp up load and see where the system breaks.
>>
>>  Also, are you running a tool like linux "sar" on each server at 1 second
>> sample rates (since your test is only lasting 1 second (rate = 15000,
>> num_connections = 15000))?  This is critical to have this data so you can
>> tell what the resource utilization is on each of your servers (proxy and
>> web server AND load generator).
>>
>>    Tim****
>>
>> ** **
>>
>> Senior Systems Performance Engineer****
>>
>> Synacor, Inc.****
>>
>> 40 La Riviere Drive, Suite 300****
>>
>> Buffalo, NY 14202****
>>
>> 716-362-3444
>>
>>   From: Vikash Kumar <vikash.kumar at oneconvergence.com>
>> Date: Monday, October 22, 2012 7:57 AM
>> To: "httperf at linux.hpl.hp.com" <httperf at linux.hpl.hp.com>
>> Subject: [httperf] Re: Connection rate is very low
>>
>>    I am also getting TCP SYN flooding msg in dmesg of *Haproxy host m/c*:
>>
>>   [ 2965.742025] possible SYN flooding on port 80. Sending cookies.
>>
>>    Command   *cat /proc/sys/net/ipv4/tcp_syncookies is set to 1.*
>> *
>> *
>> *   *What can be done to stoop this message?
>> *
>> *
>> Regards,
>> Vikash* *
>>
>> On Mon, Oct 22, 2012 at 5:13 PM, Vikash Kumar <
>> vikash.kumar at oneconvergence.com> wrote:
>>
>>> Hi all,
>>>
>>>     I am doing httperf test but I am getting very less *connection rate.
>>> * Below is the command and O/P of the command:
>>>
>>>   httperf --hog --client=0/1 --server=39.0.0.1 --port=80 --uri=/1kb
>>> --rate=15000 --send-buffer=4096 --recv-buffer=16384 --num-conns=15000
>>> --num-calls=1
>>> Maximum connect burst length: 26
>>>
>>>  Total: connections 15000 requests 9614 replies 9589 test-duration
>>> 39.149 s
>>>
>>>  Connection rate: 383.1 conn/s (2.6 ms/conn, <=12651 concurrent
>>> connections)
>>> Connection time [ms]: min 2.8 avg 4765.6 max 38926.4 median 3202.5 stddev
>>> 4236.4
>>> Connection time [ms]: connect 9995.9
>>> Connection length [replies/conn]: 1.000
>>>
>>>  Request rate: 245.6 req/s (4.1 ms/req)
>>> Request size [B]: 64.0
>>>
>>>  Reply rate [replies/s]: min 0.0 avg 273.8 max 1157.2 stddev 430.1 (7
>>> samples)
>>> Reply time [ms]: response 931.1 transfer 0.2
>>> Reply size [B]: header 174.0 content 733.0 footer 0.0 (total 907.0)
>>> Reply status: 1xx=0 2xx=6553 3xx=0 4xx=35 5xx=3001
>>>
>>>  CPU time [s]: user 12.21 system 26.94 (user 31.2% system 68.8% total
>>> 100.0%)
>>> Net I/O: 232.5 KB/s (1.9*10^6 bps)
>>>
>>>  Errors: total 5411 client-timo 0 socket-timo 5386 connrefused 0
>>> connreset 25
>>> Errors: fd-unavail 0 addrunavail 0 ftab-full 0 other 0
>>>
>>>
>>>    All the request are going through *Haproxy. * Configuration of the
>>> Haproxy host m/c is:
>>>    OS : Ubuntu 10.04
>>>    Model name: Intel(R) Xeon(R) CPU E5504  @ 2.00GHz
>>>    HA-Proxy version 1.4.22
>>>    No.of CPU's : 8
>>>
>>>     My server and client m/c's are 8 cores. I am using *lighttpd* as
>>> server.
>>>
>>>     My FD_LIMIT is 400000.
>>>    I had increased  number of sockets at optimum figure.
>>>
>>>
>>>     Why I am getting such a small *conn. rate, req. rate and reply rate?
>>>  *
>>> *
>>> *
>>> *  **  *When the same test is done for back- to back connection ie.
>>> without Haproxy, then I am able to achieve 27K *conn. rate, req. rate
>>> and reply rate*.
>>>
>>>   Can  anybody help me with this?
>>>
>>>  Thanx,
>>> Vikash
>>>
>>>
>>>
>>
>
-------------- next part --------------
_______________________________________________
httperf mailing list
httperf at linux.hpl.hp.com
http://www.hpl.hp.com/hosted/linux/mail-archives/httperf/


More information about the httperf mailing list