When you say "virtual environments" what do you mean? 

With "containers" or Solaris zones you have a very lightweight virtualized 
runtime environment where each container uses teh same OS kernel, with 
separation being really just a chroot combined with a cgroup. Performance 
tuning is no different than a native environment.

With a hypervisor setup (Xen, Virtual Box, etc) you have the coarse-grained 
Xen scheduler taking the place of the OS scheduler. This is great 
mainstream technology. Electronic trading is not a mainstream use case. It 
shouldn't be a surprise that a mainstream platform doesn't fit a low 
latency use case. 

My two cents.

Peter


 


On Tuesday, March 28, 2017 at 8:46:03 AM UTC-4, Erik Svensson wrote:
>
> Howdy guys! 
>
> My journey into mystification with virtual environments continues. 
>
> We have a number of market data feeds from a number of markets. Each 
> market feed is running on its own virtual host. 
> I’m keeping an eye of gc logs and such. 
> One thing I noticed after a while is that kernel time seems to grow over 
> time, eventually exceeding user time. If you reboot the machine it goes 
> back to ‘normal’ and then starts over again. 
> I have never seen this behaviour with physical hw. 
>
> Does this ring a bell with anyone? Right now I’m thinking about rebooting 
> the virtual hosts every weekend but that is an unsatisfactory solution. 
>
> cheers 
> Erik

-- 
You received this message because you are subscribed to the Google Groups 
"mechanical-sympathy" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to [email protected].
For more options, visit https://groups.google.com/d/optout.

Reply via email to