|
On Mon, 2003-06-23 at 21:50, Roberto Nibali wrote:
> > With this and 4 real servers I was able to get over 108000 connections
> > through the load balancer to the real servers. At this point the client
> > machines started dying :)
>
> In which ways? Load, file handles, memory, interrupt congestion?
I had 2, the 2.4 kernel machine just froze. The 2.5 machine just slowed
to the point where I could not use a local console to find out what was
giving out however from the sounds from the hdd I would guess it was
swapping a lot. It is not something that bothers me too much.
> > my kid sis can write perl scripts that are more useful for this
>
> Yes, if you only need to get a TCP connection, you're right, once you
> need to handle some of the html data, you might need to delve deeper
> into perl programming.
Agreed, this is something I want to work on as well (in c of course,
perl would not be too useful for a serious benchmarking proggie), while
it will do a lot less than ab it should be able to push harder.
> > all docs on lvs performance are out of date, to this end I need to write
> > a paper on this
>
> Ok, what exactly is out of date on them? I'm simply asking so we can add
> a note where appropriate.
2.2 kernels, no mention of how 2.5 will do, even 2.4 was in it's early
days and the amount of connections the machine could take, if I can push
above this then I am sure it could go a lot higher. While the test
conditions were most excellent, esp with the amount of hardware they had
it would have been neat if they had used something that could take more
connections on the backend as well. This is something I want to spend
more time working on.
While it is nice to see how far you can push web server x when doing
load balancer testing it is really not relevent, you want to know how
much the lvs machine can take, how many connections and what throughput
so the best way to test this is to have a dedicated client on the real
servers that will just accept connections.
Also it would be nice to see how the 1Gbps cards use pci-x and what
bandwidth can be put through them. Also it would be nice to see the
10Gbps cards. This is something I am going to try and do at work
because in out situation they would like to know it will be able to
handle any load.
> > If you have any further questions please feel free to email me. I will
> > post once I have done more benchmarking and written docs about it.
>
> ... I hope you will include the numbers I requested 3 times in your
> document. Other than that we'll certainly appreciate some documentation
> or paper on the performance of LVS.
:) Sorry about that. The numbers you requested.
Time per request: 2.00 [ms] (mean)
Document Length: 5 bytes
If you would like anything more please let me know.
I hope my last email did not sound too grouchy, I am very thankful for
all of the work you have all done and you have all been more helpful.
Take care and thanks again - RL
--
MSN:lazzurs@xxxxxxxxxxxxxx |"All that is etched in stone
Yahoo:admroblaz AIM:admroblaz |is truly only scribbled in
ICQ:66324927 |sand" - RL
Jabber:admroblaz@xxxxxxxxxx |Join Eff http://www.eff.org
e-mail:lazzurs@xxxxxxxxxxxxxxxxx|Take care all - Rob Laz
|
