Re: VM benchmarks

It's really strange. I notice a huge improvement in non-virtualized environments as well.

I modeled my network on all old laptops (like sub-500mhz era) using haproxy pointed to two backend nginx servers and I get 10-30ms response for static content (client and servers all on the same 100mbit LAN). I then modeled the same setup in VirtualBox (all on the same computer) from client (host OS) to servers (three guest OS's) and I have an average time to fully downloaded content of over 150ms. And yes the CPU supports VT-x and the virtualization is configured to use it.

Watching CPU / RAM / IO on all involved nodes (in both networks) shows that there is virtually no load (htop itself is using more resources).

Is there a system call that haproxy uses that simply does not translate efficiently to a virtualized environment?

-a

On Oct 26, 2010, at 4:59 PM, Daniel Storjordet wrote:

>
> We just moved HAProxy from ESXi servers into two dedicated Atom servers.
>
> In the first setup the HAProxy innstallations balanced two webservers in the same ESXi enviorment. The web access times for this config was inbetween 120-150ms (Connect, Request, Download).
>
> In the new config the dedicated HAProxy boxes are located in a seperate datacenter 500km away from the same ESXi web servers. With this config we get lower web access times. Inbetween 110-130ms (Connect, Request, Download).
>
> I expect that also moving the web servers to the new datacenter will result in an even better results.
>
> --
> mvh.
>
> Daniel Storjordet
>
> D E S T ! N O :: Strandgata 117 :: 4307 Sandnes
> Mob 45 51 73 71 :: Tel 51 62 50 14 daniel@desti.no :: http://www.desti.no
> www.destinet.no - Webpublisering på nett
> www.func.no - Flysøk på nett
>
>
>
> On 26.10.2010 16:38, Ariel wrote:

>> Does anyone know of studies done comparing haproxy on dedicated hardware vs virtual machine? Or perhaps some virtual machine specific considerations? >> -a Received on 2010/10/27 16:58