Xen users - Dec 2012 - How to optimize pre-copy algorithm of xen to minimize downtime?

Hello,
         If I want to optimize the performance of precopy algorithm so that
live migration of virtual machine using xen occurs with minimum
downtime,then how to do it?

regards,
Digvijay


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

Hello,
         If I want to optimize the performance of precopy algorithm so that
live migration of virtual machine using xen occurs with minimum
downtime,then how to do it?

regards,
Digvijay


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

On Tue, 2012-12-11 at 14:32 +0000, digvijay chauhan
wrote:
> Hello,
>          If I want to optimize the performance of precopy algorithm so
> that live migration of virtual machine using xen occurs with minimum
> downtime,then how to do it?
Perhaps start by analysing the code and seeing if you can spot some
inefficiency which you can address?

I also recommend you talk to your course advisor and get their input.

Ian.

Hi,

there was a list of issues with migration performance published at 
http://www.slideshare.net/xen_com_mgr/vm-live-migration-speedup-in-xen - 
not sure this helps.

Otherwise, you really need to start looking at the code, run bechnmarks, 
profile, etc. There are some tools you can look at to help you, e.g.,
1) http://www.slideshare.net/xen_com_mgr/5-xs-asia11gdunlapxenalyze
2) http://wiki.xen.org/wiki/Category:Performance - there some articles 
on how to profile, collect data, etc. in this category

Regards
Lars

On 11/12/2012 14:40, Ian Campbell wrote:
> On Tue, 2012-12-11 at 14:32 +0000, digvijay chauhan wrote:
>> Hello,
>>           If I want to optimize the performance of precopy algorithm so
>> that live migration of virtual machine using xen occurs with minimum
>> downtime,then how to do it?
> Perhaps start by analysing the code and seeing if you can spot some
> inefficiency which you can address?
>
> I also recommend you talk to your course advisor and get their input.
>
> Ian.
>
>
> _______________________________________________
> Xen-users mailing list
> Xen-users@lists.xen.org
> http://lists.xen.org/xen-users

On 11/12/12 14:33, digvijay chauhan wrote:
>
> Hello,
>          If I want to optimize the performance of precopy algorithm so 
> that live migration of virtual machine using xen occurs with minimum 
> downtime,then how to do it?
Having spent quite a bit of time studying the time that a domain is down 
during migration in XenServer, I''m not convinced pre-copy is what is
the
biggest part of downtime, assuming the Open Source product behaviour is 
at least mostly doing the same things - although what the host and guest 
is up to before it gets suspended/paused will have a big impact as well. 
If the guest is VERY busy, you may well end up with very little memory 
that doesn''t need to be copied in the final copy - not sure how you can
improve this.

Have you made some measurements to show where the time is spent during 
downtime.
What workloads are you looking at? What downtime numbers have you got at 
present, and what portion of this is down to actually copy time, and 
what is other things that need to happen during migration (e.g. bringing 
the virtual HD and Net devices down, and then up again on the new guest)?

My above comments assume there is a FAST link between the old and new 
host (my testing has been primarily using "localhost" migration, so
the
new guest is on the same machine as the old one). For a slow link, you 
may find that the copy time is more of the whole time, but again, I''m 
not sure there is a huge amount that can be done without a rather large 
amount of added complexity - and that may well be better put elsewhere, 
e.g. compression on the network link, or some such.

I have patch for Linux 3.7rcX that improves the time Dom0 spends mapping 
in the guest memory, which does have a good impact on the overall copy 
time, but not that huge impact on the downtime (since the copy time is 
only small part of overall downtime, at least in my testcases). Look at 
list archives for Friday 7th of Dec for the latest version.

--
Mats
>
> regards,
> Digvijay
>

The authors of "High Performance Virtual Machine Migration with RDMA
over Modern Interconnects"  also did some changes to the pre-copy
mechanism. It''s at
http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.73.8714&rep=rep1&type=pdf
There was two things they did, change something that makes it more
efficient in the transfer ordering (don''t remember) and also bandwidth
throttling.

As *lol* almost normal for research about speeding up stuff, the
authors did not bother to give a link to the sources for what they
changed.
Anyway, if you need to dive into the topic. it''s relevant reading. No
matter if you use RDMA.

HTH,
Florian

2012/12/12 Lars Kurth <lars.kurth@xen.org>:
> Hi,
>
> there was a list of issues with migration performance published at
> http://www.slideshare.net/xen_com_mgr/vm-live-migration-speedup-in-xen -
not
> sure this helps.
>
> Otherwise, you really need to start looking at the code, run bechnmarks,
> profile, etc. There are some tools you can look at to help you, e.g.,
> 1) http://www.slideshare.net/xen_com_mgr/5-xs-asia11gdunlapxenalyze
> 2) http://wiki.xen.org/wiki/Category:Performance - there some articles on
> how to profile, collect data, etc. in this category
>
> Regards
> Lars
>
>
> On 11/12/2012 14:40, Ian Campbell wrote:
>>
>> On Tue, 2012-12-11 at 14:32 +0000, digvijay chauhan wrote:
>>>
>>> Hello,
>>>           If I want to optimize the performance of precopy
algorithm so
>>> that live migration of virtual machine using xen occurs with
minimum
>>> downtime,then how to do it?
>>
>> Perhaps start by analysing the code and seeing if you can spot some
>> inefficiency which you can address?
>>
>> I also recommend you talk to your course advisor and get their input.
>>
>> Ian.
>>
>>
>> _______________________________________________
>> Xen-users mailing list
>> Xen-users@lists.xen.org
>> http://lists.xen.org/xen-users
>
>
>
> _______________________________________________
> Xen-users mailing list
> Xen-users@lists.xen.org
> http://lists.xen.org/xen-users


-- 
the purpose of libvirt is to provide an abstraction layer hiding all
xen features added since 2006 until they were finally understood and
copied by the kvm devs.