httpress benchmark utility now supports SSL via GNUTLS

Yaroslav yarosla at gmail.com
Tue Jan 17 18:58:42 CET 2012


The difference is huge. Here are some results.

Running tests against nxweb server, minimal c handler '<p>Hello,
world!</p>' (20 bytes).
Ubuntu 11.10 x64, 4 core CPU without AES-NI support.

Testing cipher suite:
ab info: TLSv1/SSLv3,ECDHE-RSA-AES256-SHA,2048,256
httpress info: ECDHE_RSA_AES_256_CBC_SHA1
- Key Exchange: ECDHE-RSA
- Ephemeral ECDH using curve SECP256R1
- Protocol: TLS1.0
- Certificate Type: X.509
- Compression: NULL
- Cipher: AES-256-CBC
- MAC: SHA1

Non-keep-alive:

ab -c 100 -n 6000
result: 380 rps

httpress -c100 -n6000 -t4 -z
'NORMAL:-CIPHER-ALL:+AES-256-CBC:-VERS-TLS-ALL:+VERS-TLS1.0'
result: 490 rps

siege -c 100 -r 60 (setup cipher via siegerc: ECDHE-RSA-AES256-SHA)
result: 500 rps

Keep-alive:

ab -c 100 -n 60000 -k
result: 22 000 rps

httpress -c100 -n6000 -t4 -z
'NORMAL:-CIPHER-ALL:+AES-256-CBC:-VERS-TLS-ALL:+VERS-TLS1.0'
result: 42 000 rps

siege -c 100 -r 600 (setup cipher via siegerc: ECDHE-RSA-AES256-SHA)
result: unable to complete the test due to excessive failures; tried to
test against nginx - same result

Plain http test (non-keep-alive):

siege -c 100 -r 1000
result: 11 000 rps

ab -c 100 -n 100000
result: 16 000 rps

httpress -c 100 -n 100000 -t4
result: 44 000 rps

Plain http test (keep-alive):

siege -c 100 -r 4000
result: 25 000 rps

ab -c 100 -n 400000 -k
result: 71 000 rps

httpress -c 100 -n 1000000 -t4 -k
result: 180 000 rps


On Tue, Jan 17, 2012 at 6:16 PM, Nikos Mavrogiannopoulos <nmav at gnutls.org>wrote:

> On Tue, Jan 17, 2012 at 1:10 AM, Yaroslav <yarosla at gmail.com> wrote:
> > Hi,
> > I thought this might be interesting for some people on this list.
> > I'd like to announce addition of SSL support to
> > httpress: https://bitbucket.org/yarosla/httpress/
> > Compared to ApacheBench and siege httpress offers greatly improved
> > performance. As well as precise cipher suite selection via command line
> > option.
>
> Interesting. Did you notice differences in the web servers performance
> comparison by using operations in parallel in httpress?
>
> regards,
> Nikos
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: </pipermail/attachments/20120117/1104ee73/attachment.htm>


More information about the Gnutls-devel mailing list