Re: [Xen-devel] Taking on a Xen development project

[Andrew Cooper <andrew.cooper3@xxxxxxxxxx>]

On 12/12/2015 22:30, Joshua Otto wrote:
> On Sat, Dec 12, 2015 at 10:19:31AM +0800, Yang Hongyang wrote:
>> On 2015å12æ11æ 01:23, Andrew Cooper wrote:
>>> Hello - thankyou for your interest.
>>>
>>> One area to look at might be the parameters to the live migration
>>> looping.  As part of the migration v2 rework I did in the 4.6 dev
>>> period, I left all of that alone, and it is in a working but poor state.
>>>
>>> In the past, there have been several research investigations into
>>> improving the live migration algorithm, such as tracking the rate of
>>> dirtying of memory, or attempting to resume the domain on the far side
>>> and fault the final memory across.
>> I think you mean postcopy here? The hypervisor then needs to maintain
>> a dirty page bitmap and generate pagefault when a page is not yet
>> tranferred to the far end.
>> This feature already merged into QEMU2.5(kvm patch which generates
>> pagefault also been merged into linux kernel mainline), if you want a
>> reference, you can take a look at those patches.
>> This surely is a great aera to work on.
>>
>>> If you are interested in perusing this, start with reading
>>> docs/features/migration.pandoc in the Xen tree.
> We'd definitely be interested in working on live migration!  The feature
> is essentially an implementation of the approach described in Section
> 5.1 of [0], right?

Yes - Section 5.1 is quite a good general description of live migration
(even after my white-room redesign from first principles), although be
aware that some of the more technical details are now out of date.

>
> Would the focus of the project be to implement and evaluate postcopy
> live migration in Xen, then, or to more generally build on previous
> research efforts? (either way sounds like fun!)

I hadn't really though that far ahead.  I was more suggestion that the
general area of live migration has a lot in the way to offer for
projects, be it implementing someone else's research, or researching a
new area yourself.

There are definitely areas where improvements can be made, and some of
these would be an easier introduction to the codebase than to start with
a full postcopy implementation.  (Not that I wish to put you off
postcopy, but it does come with a number of non-trivial problems to
solve as part of getting the scaffolding in place.)

I guess it depends on what you are looking to get out of a project like
this, and what timescale you have.

~Andrew

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel