RE: [Xen-devel] More about Resource Scheduling and Accounting

["Rolf Neugebauer" <rn@xxxxxxx>]

> -----Original Message-----
> From: xen-devel-admin@xxxxxxxxxxxxxxxxxxxxx [mailto:xen-devel-
> admin@xxxxxxxxxxxxxxxxxxxxx] On Behalf Of cs0u210a@xxxxxxxxxxxxxxx
> Sent: Friday, October 10, 2003 10:31 PM
> To: xen-devel@xxxxxxxxxxxxxxxxxxxxx
> Cc: Keir.Fraser@xxxxxxxxxxxx; Ian.Pratt@xxxxxxxxxxxx
> Subject: [Xen-devel] More about Resource Scheduling and Accounting
>
> Hi,
>
>     Sorry I have been busy with some academic issues since my last email
> but
> should now focus on Xen with all my time.
>
>     As stated before I would like to work on the Resource Scheduling and
> Accounting part (in addition, try to relate a bit to the OS
> suspend/resume for
> migration), which at the same time will be my honour year project. The
> functions/requirements that Ian has mentioned before is still at a high
> level.
> Hence I would like to know if there are some specific requirements (eg
> standard
> of interface to other kernel functions and interface for domain0 etc)
> that you
> would like to mention, especially by the Xen team. I welcome any
> comments/suggestions by anyone in the Xen-devel maillist as well, be it
> additional features, concepts or designs or anything else.
>
>     A particular question I would like to know is how well the BVT
> algorithm is
> functioning, and what is the weakness at the moment. In the SOSP 2003
> paper,
> 3.3.1 CPU Scheduling mentioned other scheduling algorithms could be
> trivially
> implemented over the generic abstraction and per-domain scheduling
> parameters
> can be adjusted....in Domain0. First, could you tell me all files related
> to
> the scheduling algorithm and the abstraction, this would greatly help me
> to
> start. Second, is it true that per-domain parameters can be ajusted by
> Domain0?
> How well is this performing? What are the things need to be improved?

in xen the scheduler is implemented in common/schedule.c. sched.h includes
the task structure and common ac_timer.c provides a timer interface. The
interface to changes scheduling parameters is exported to dom via
common/dom0_ops.c and in via a hypercall (HYPERCALL_dom0_op). See
xi_sched_domain.c for an example.

A new scheduler mainly requires changes to schedule.c but also to
dom0_ops.c to change the parameters passed from domain0.

BVT provides prop. fairness. That works fine. BVT also has a mechanism to
decrease wakeup latencies (warping). This currently happens if a domain
receives an event. As a domain in the current version receives a periodic
ticker interrupt this means it gets warped periodically (subject to the
limits provided by BVT). This also means that if a more important event
arrives (e.g., network packet) a domain might not be warped. This may be a
problem. I have patches which hopefully solve this (under test at the
moment). These patches disable the periodic ticker and on the xenolinux
side are based on-demand timer patches IBM provides for the s390 port of
Linux.

I'd suggest to try to implement one or more different scheduler, maybe one
offering absolute shares, rather than prop. fairness. As we say in the
paper this should be quite straight forward.

Another limitation of the current scheduler is that it doesn't provide
migration of domains between CPUs. A domain is currently assigned a fixed
CPU. migrating it to others CPUs when that CPU is busy would be useful.
Maybe you want to look at "take scheduling" (sorry don't have a reference
handy)

for network scheduling see Keir's recent checkin.

Rolf


>
>     One more thing, I wonder if you can provide me some pointers to other
> materials you think will be useful for the work such as research papers,
> texts,
> how-to, source codes etc.
>
> Many thanks,
>
> Yan-Ching CHU
>
>
>
>
>
>
>
> -------------------------------------------------------
> This SF.net email is sponsored by: SF.net Giveback Program.
> SourceForge.net hosts over 70,000 Open Source Projects.
> See the people who have HELPED US provide better services:
> Click here: http://sourceforge.net/supporters.php
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@xxxxxxxxxxxxxxxxxxxxx
> https://lists.sourceforge.net/lists/listinfo/xen-devel




-------------------------------------------------------
This SF.net email is sponsored by: SF.net Giveback Program.
SourceForge.net hosts over 70,000 Open Source Projects.
See the people who have HELPED US provide better services:
Click here: http://sourceforge.net/supporters.php
_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxxxx
https://lists.sourceforge.net/lists/listinfo/xen-devel