[httperf] Missing requests and connection resets with httperf and squid

[Dave Jenkins]

Hi Martin,

Thanks, that makes sense, though I can't access the patch. Some
further findings...

When viewing the squid+apache site in a browser, the response headers include:

Connection:keep-alive

However when using httperf, I get:

Connection: close

in each response.

I suspect the issue is that Squid apparently does not fully support
HTTP/1.1, see e.g.
http://serverfault.com/questions/58803/how-to-configure-squid-to-send-back-a-http-1-1-response-for-a-http-1-1-request
.

I tried the 'http11' option to squid's http_port config:

#          http11       Enables HTTP/1.1 support to clients. The HTTP/1.1
#                       support is still incomplete with an internal HTTP/1.0
#                       hop, but should work with most clients. The main
#                       HTTP/1.1 features missing due to this is forwarding
#                       of requests using chunked transfer encoding (results
#                       in 411) and forwarding of 1xx responses (silently
#                       dropped)

This seems to have sorted it out: I now get:

Total: connections 3 requests 100 replies 100 test-duration 12.164 s
Errors: total 0 client-timo 0 socket-timo 0 connrefused 0 connreset 0
Errors: fd-unavail 0 addrunavail 0 ftab-full 0 other 0

So I can now do some benchmarking. :)

Thanks,

Dave


On 10 December 2010 16:21, Arlitt, Martin <martin.arlitt at hp.com> wrote:
> Hi Dave
>
> I expect the issue in the second test is that persistent connections are not supported, but httperf is not recognizing the Connection: close header from the proxy.
>
> One solution is described in a recent thread on the mailing list:
>
> http://www.hpl.hp.com/hosted/linux/mail-archives/httperf/2010-November/thread.html
>
> specifically, you can patch httperf to close the connection if so instructed by the server:
>
> http://www.hpl.hp.com/hosted/linux/mail-archives/httperf/2010-November/000677.html
>
> thanks
> Martin