Willy Tarreau a écrit :
Hello,
On Fri, Jun 25, 2010 at 05:15:02PM +0200, eni-urgence wrote:
Hello
Is anyone have already notice a perfomance decrease with and
without haproxy ? I m' the only one on the proxy and web server, and
the http resquest is served 0.5 and 1s slower than without haproxy?
How are you measuring the delay ? From haproxy's logs or from the
client ? 1 second is huge. A typical traversal delay should be in
the range of about 40-100 microseconds, or 10-25000 times lower
than what you're observing. Maybe you're having a config issue,
or maybe there's something wrong on the machine you're running
it on (eg: is it virtualized, this is causing new classes of hard
to debug trouble) ?
Regards,
Willy
Hello Willy and thanks for your help and interrest.
It's an test embeded in the php/ajax application. According to the
developper, the measure is made like that :
for one ajax call, he creates a timer on the call. Then at the beginning
of the php scripts, he creates an other time which evaluate the
generation time of the script and call back it to ajax.
On client side, when the response is received, it gives the transfer
time+the php time. Then he substracts the php time (return by ajax) from
the total duration . This duration include the transfert time of the
request to the server.
I had tested transfer between a web server in production (and quite
loaded ~1 ) and the proxy machine (not so powerfull then the webserver).
The network path for the test is like that :
Client (on LAN) <------> Firewall <------> stunnel on VIP
192.168.100.153 (on the proxies) <------> Haproxy (listening on
127.0.0.1) <------> WebServer (listening on 192.168.100.41) (on the
proxy too for the moment )
The proxy server is a Quad core Intel(R) Xeon(R) CPU X5460
3.16GHz with 4G RAM. Os is CentOS 5.5 64b. Not virtualised. I haven't
tweak the OS parameters according to your advice (found on many
sites). With one RAID1.
The webServer, which serves for measurement reference, is a double Quad
Core with 6Go RAM and a double RAID1 (one for system, one for php pages
and some data)
After some test, it appears that is a hardware/design problem. I made a
new virtual host (https) on the proxy machine, call it directly not
through haproxy and the transfer time are similar. Now i had to found
why . Is think the double RAID and the IO disk is for much in those
difference.
Thanks for your help and time. And sorry for inconvenience.
NICOLE Emerik
Newbie french user of Haproxy
