[httperf] zero results in reply rate

[Arlitt, Martin]

Hi Roman

Have you tried tests that have lasted longer than 4.998 seconds?

Looking at the output you sent previously, at the end of the "Reply rate" line you will see "(0 samples)". I believe this is why the reply statistics are all zero. Samples are collected over windows of RATE_INTERVAL seconds; RATE_INTERVAL is defined in httperf.c as 5.0. The test results you sent had a duration of 4.998s, which would explain why the "Reply rate" line says 0 samples. In stat/basic.c (dump function) the reply statistics are only calculated if there is at least one sample.

Thanks
Martin


-----Original Message-----
From: Roman Florea [mailto:roman.florea at tut.fi] 
Sent: Friday, April 12, 2013 10:35 AM
To: Arlitt, Martin
Cc: httperf at linux.hpl.hp.com
Subject: RE: [httperf] zero results in reply rate

Hi Martin,

thanks for your attention.
Yes it is just a header... the code is a slightly edited version of sample libevent httpd server https://github.com/libevent/libevent/blob/master/sample/http-server.c
I have tried a quick patch to add some bytes in reply but it did not help.

Roman.


Quoting "Arlitt, Martin" <martin.arlitt at hp.com>:

> Hi Roman
>
> Does the reply consist only of a header (and no body)?
>
> Thanks
> Martin
>
>
> -----Original Message-----
> From: httperf-bounces at linux.hpl.hp.com 
> [mailto:httperf-bounces at linux.hpl.hp.com] On Behalf Of Roman Florea
> Sent: Friday, April 12, 2013 7:34 AM
> To: httperf at linux.hpl.hp.com
> Subject: [httperf] zero results in reply rate
>
> Hello,
>
> while using httperf for benchmarking a very simple http server, during 
> some experiments it returns 0 values for the reply rates though all 
> replies were received and no errors reported.
> And it happens both on low values and high requested values.
> Example output for requested conn rate of 360
>
>
> httperf --hog -v  --server target --uri "/dump" --num-conn 1800  
> --num-call 1 --timeout 5 --rate 360 --port 5000 httperf --verbose  
> --hog --timeout=5 --client=0/, what are those 0 samples., what are  
> those 0 samples.1 --server=target --port=5000 --uri=/dump --rate=360  
> --send-buffer=4096 --recv-buffer=16384
> --num-conns=1800 --num-calls=1
> httperf: maximum number of open descriptors = 65000 Maximum connect  
> burst length: 1
>
> Total: connections 1800 requests 1800 replies 1800 test-duration 4.998 s
>
> Connection rate: 360.1 conn/s (2.8 ms/conn, <=2 concurrent  
> connections) Connection time [ms]: min 0.4 avg 0.5 max 4.3 median  
> 0.5 stddev 0.1 Connection time [ms]: connect 0.2 Connection length  
> [replies/conn]: 1.000
>
> Request rate: 360.1 req/s (2.8 ms/req)
> Request size [B]: 84.0
>
> Reply rate [replies/s]: min 0.0 avg 0.0 max 0.0 stddev 0.0 (0  
> samples) Reply time [ms]: response 0.3 transfer 0.0 Reply size [B]:  
> header 120.0 content 0.0 footer 0.0 (total 120.0) Reply status:  
> 1xx=0 2xx=1800 3xx=0 4xx=0 5xx=0
>
> CPU time [s]: user 3.47 system 1.52 (user 69.5% system 30.4% total  
> 99.9%) Net I/O: 71.7 KB/s (0.6*10^6 bps)
>
> Errors: total 0 client-timo 0 socket-timo 0 connrefused 0 connreset 0
> Errors: fd-unavail 0 addrunavail 0 ftab-full 0 other 0
>
>
> This happens on certain connection rates, and also happens for  
> different server and client OSes. The server code is using libevent  
> and simply returns a 200 header for every connection it gets. The  
> issues is seen on autobench progress:
> dem_req_rate	req_rate_select	con_rate_select	min_rep_rate_select	avg_rep_rate_select	max_rep_rate_select	stddev_rep_rate_select	resp_time_selectnet_io_select	errors_select
> 32500	32267.4	322.7	32420	32420	32420	0	0.475	6428.2	0
> 35000	34694.3	346.8	34816.4	34816.4	34816.4	0	0.7375	6911.9	0
> 37500	37177.2	371.6	0	0	0	0	0.9	7406.4	0
> 40000	36438.7	364.2	31454.1	31454.1	31454.1	0	3.3625	7259.2	0
> 42500	23170.5	231.9	22745.4	22745.4	22745.4	0	25.75	4616.1	0
>
> The other parameters seem fair, and error rate is always 0.
> Could someone, please, explain what do these 0 mean and why this  
> might happen.
>
> Thank you.
>
> Kind regards,
> Roman Florea.
>
>
>
>
> _______________________________________________
> httperf mailing list
> httperf at linux.hpl.hp.com
> http://www.hpl.hp.com/hosted/linux/mail-archives/httperf/
>