Re: [Xen-devel] Serious issues with xenpaging

[Andres Lagar-Cavilla <andreslc@xxxxxxxxxxxxxx>]

On Jan 3, 2014, at 1:41 PM, Konrad Rzeszutek Wilk <konrad.wilk@xxxxxxxxxx> 
wrote:

> On Fri, Jan 03, 2014 at 09:49:36AM -0500, Andres Lagar-Cavilla wrote:
>> 
>> On Dec 31, 2013, at 11:31 AM, Tim Deegan <tim@xxxxxxx> wrote:
>> 
>>> At 10:33 -0500 on 31 Dec (1388482410), Konrad Rzeszutek Wilk wrote:
>>>> On Mon, Dec 23, 2013 at 06:34:55PM +0000, Russell Pavlicek wrote:
>>>>> On Twitter, Florian Heigl sent a out a few messages about issues with 
>>>>> xenpaging:
>>>>> 
>>>>> ---
>>>>> 19-Dec: Anyone successfully use 
>>>>> #xen<https://twitter.com/search?q=%23xen&src=hash> 
>>>>> #xenpaging<https://twitter.com/search?q=%23xenpaging&src=hash>? docs are 
>>>>> at SLES manual, rest is mostly this: 
>>>>> http://www.gossamer-threads.com/lists/xen/devel/255798<http://t.co/P36VdL84Et>
>>>>>  dead feature or usable?
>>>>> 
>>>>> 22-Dec: @lars_kurth<https://twitter.com/lars_kurth> 
>>>>> @RCPavlicek<https://twitter.com/RCPavlicek> Hey guys, I wrote down as 
>>>>> much as I could https://piratenpad.de/p/Ik3lOBLniq1L5TEM   
>>>>> <https://t.co/e5LQCUD9d0> (since I'm on holiday and not constant online)
>>>>> 
>>>>> 22-Dec: Yay, tested #xen<https://twitter.com/search?q=%23xen&src=hash> 
>>>>> Xenpaging (memory overcommit)
>>>>> [x] largely untested
>>>>> [x] docs outdated
>>>>> [x] syntax+logic changed
>>>>> [x] broken
>>>>> ---
>>>>> 
>>>>> [I've taken the liberty of removing the colorful expletive from the final 
>>>>> post]
>>>>> 
>>>>> Is Florian's assessment correct, or is there somewhere we can point him 
>>>>> for help?  I'm on vacation this week, but if someone replies to me, I 
>>>>> will try to forward the information appropriately.
>>>> 
>>>> The Maintainers file implies otherwise. Let me CC the maintainers.
>>> 
>>> Andres really owns this code, so I'll punt to him for an official
>>> answer, but:
>> The part actively maintained is the hypervisor support for paging, and the 
>> interface.
>> 
>> tools/xenpaging is one way to consume that interface. It seems to have 
>> suffered from bitrot.
> 
> What is the other interface? Thanks!

Not sure what the question is. There is one interface. What I was referring to, 
is that tools/xenpaging implements one specific paging policy: victim 
selection, rate limiting, paging target, all of these are algorithms that 
entirely define what bang for your money you will get.

Andres
>> 
>> So other than echoing Tim's points below, I'll add
>> 
>> - Some interesting ideas thrown around by Florian in his notes. Could lead 
>> to a robust discussion in xen-devel … if Florian is still interested.
>> 
>> - Perhaps the developers who are interested (myself included) should make a 
>> decent effort at improving the in-tree tools. There is the argument that for 
>> example KSM gives KVM users a sharing solution that just works, whether you 
>> like the results or not. In that vein xenpaging apparently doesn't cut it, 
>> nor the absence of a basic sharing tool.
>> 
>> One simple paging tool could be lazy restore. There is some interest out 
>> there, it would be relatively straightforward to codify.
>> 
>> Andres
>>> 
>>> - It's been listed as a 'tech preview' on the feature list since it went
>>> in.  http://wiki.xenproject.org/wiki/Xen_Release_Features says:
>>> "Preview, due to limited tools support. Hypervisor side in good shape."
>>> 
>>> - I can't say anything about SuSE's apparent support for it, except
>>> that ISTR Olaf worked at/for/with SuSE at the time.
>>> 
>>> - Patches would, of course, be welcome.
>>> 
>>> Tim.
>> 


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