Re: [Vserver] vserver build docs, and vserver docs in general

From: Herbert Poetzl <herbert_at_13thfloor.at>
Date: Sun 21 May 2006 - 16:00:03 BST
Message-ID: <20060521150003.GA7426@MAIL.13thfloor.at>

On Sun, May 21, 2006 at 12:11:11PM +0200, ADNET Ghislain wrote:
> ok thanks Herbert, i continue the batch of question then,
>
> I have seen in a faq that to have better fair share of disk IO between
> guest this one can be usefull:
>
> ???? CFQ I/O scheduler ???
> ?
> ? The CFQ I/O scheduler tries to distribute bandwidth equally
> ? among all processes in the system. It should provide a fair
> ? working environment, suitable for desktop systems.
> ?
> ? Symbol: IOSCHED_CFQ [=y]
> ? Prompt: CFQ I/O scheduler
> ? Defined at block/Kconfig.iosched:33
> ? Location:
> ? -> Block layer
> ? -> IO Schedulers
>

yes, in recent kernels, the CFQ scheduler acts
as a per guest scheduler (i.e. it has a scheduler
queue for each guest)

> what do you know about it , is it usefull and be better for the vservers
> or the vserver patch allready takes care of this ?

that really depends on the requirements and the
setup, for optimal performance you have to test
with a given system/setup

> In general, can we consider that a vserver system needs to be configured
> in latency more like a desktop (many kernel cpu and io scheduling speak
> about server/desktop) so that one guest does not steal too much of the
> cpu/io of the system or it does not have any real impact for you ?
>
> Now for fair CPU scheduling we rely on this one right ?
>
> ? [*] Enable Hard CPU Limits
> ? ? [*] Limit the IDLE task

no, for 'fair' scheduling in the classical sense
you have to enable the 'skip idle time' option
(which is only available in recent devel kernels)

the hard cpu scheduler itself guarantees an upper
limit to the cpu amount consumed by a guest

> of the vserver part, your last comment speak about keeping default
> choice but this HARD cpu limit seems not to be the default (from what
> i recall). What part of the kernel have impact on the CPU sharing
> between guests ? I try to see what part make sense to have a good
> separate scheduling between guests in cpu also.

well, by default Linux-VServer aims at performance
and in this case, whole system performance, of
course this will be higher without any cpu restrains
as the tasks will be optimally scheduled

> Of course i will ask you the same about network bandwidth sharing but
> i really do not see how to do that here. I found one script on the
> how-to list but there is no real explanation there. I guess QoS must
> be put here a little like for the disk IO but really i have no idea :)
> Do you think http://www.lartc.org/ is the way to go to provide fair
> share of network IO, it seems Stochastic Fairness Queueing should be
> quite good at that :)

yes, that's the way to do it (similar to a normal
linux system), ip based bandwidth management

HTH,
Herbert

> Cordialement,
> Ghislain ADNET.
>
_______________________________________________
Vserver mailing list
Vserver@list.linux-vserver.org
http://list.linux-vserver.org/mailman/listinfo/vserver
Received on Sun May 21 16:00:43 2006

[Next/Previous Months] [Main vserver Project Homepage] [Howto Subscribe/Unsubscribe] [Paul Sladen's vserver stuff]
Generated on Sun 21 May 2006 - 16:00:50 BST by hypermail 2.1.8