[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Xen-devel] Measuring migration downtime

  • To: "Jacob Gorm Hansen" <jacobg@xxxxxxx>
  • From: "Arun Babu" <arunbabu.n@xxxxxxxxx>
  • Date: Tue, 13 Jun 2006 19:47:21 -0400
  • Cc: xen-devel@xxxxxxxxxxxxxxxxxxx
  • Delivery-date: Tue, 13 Jun 2006 16:47:43 -0700
  • Domainkey-signature: a=rsa-sha1; q=dns; c=nofws; s=beta; d=gmail.com; h=received:message-id:date:from:to:subject:cc:in-reply-to:mime-version:content-type:references; b=AgMRXLJ7pJ20yMQKAIcNmaaCBX5FvZL7DTQeJUdniyUjt9fgH5DzepKbEbOq3nfmpLCIv07QT9npLeG9Y+qpYevkSMtnemoO7728Mfx9oLxBKQuIcJWb/EmklgWsCosbOTSrgXa+pctEgXW/lcFQ0Wnw6HscacePdcgQl79Mveo=
  • List-id: Xen developer discussion <xen-devel.lists.xensource.com>

 Thanks for replying.
 I did read your paper, which gives the results in detail .
 Infact I need to do some tests exactly like the ones in your paper..(but not in such detail)
 The method discussed estimates downtime with the idea of writable working set & information on
 number of pages dirtied and hence evaluating the count of dirty pages to be transfered
 during the final stop-copy phase.
 I am quite new to xen and am not sure how much effort need to be put in to
 get information on shadow page table, dirty page count etc to repeat the tests
 you have performed ... Please let me know if we have more detailed information
 on the method of measuring this!

 I am currently trying to estimate is based on a simple socket application running on
 the guest domain, which does an infinite connect()-close() to a service running on original host. When it migrates
 to the destination, it wont be able to access the service.(i hav my n/w that way) I am using the time of last successful
 connect() call and first error in connect() to estimate downtime.(approximately)
 This gives finer values than the ones you suggested - tcpdump with floodping
  But I have a problem automating in my method: If migration happens exactly after connect() call and
  gettimeofday() of a successful/failed connect, then no longer the last succesful connect/first failed connect
  logic holds. (But always i can manually calculate with the dumps of timestamps)
 It would be of great help if you can suggest a better method!

Sorry for a long mail!


On 6/11/06, Jacob Gorm Hansen <jacobg@xxxxxxx> wrote:
What about the methods and results in our NSDI 2005 paper, don't they
tell you what you need?

You can use in tcpdump on an independent machine and log timestamps of
packets send from the migrating domain's IP, in response to fx a

You can also sync the two machine's clocks using NTP and instrument
the tools to measure timestamp of domain suspend resp. resume.


On 6/9/06, Arun Babu <arunbabu.n@xxxxxxxxx> wrote:
> Hi,
>  I need to get the exact downtime for doing a live migration with ms
> precision.
>  (I am not bothered about whole migration time which is close to 1 minute..
> I need to measure
>  the live migration downtime during which the DomU is out of service.. the
> figure quoted in the xen
>  documentation says it as ~60 ms..  )
>  I hope someone must have tried to measure this before.
Xen-devel mailing list



Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.