qemu-devel
[Top][All Lists]
Advanced

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

Re: [Qemu-devel] For 2.0? Re: [PATCH v5 00/10] migration: Optimizate the


From: Dr. David Alan Gilbert
Subject: Re: [Qemu-devel] For 2.0? Re: [PATCH v5 00/10] migration: Optimizate the xbzrle and fix one corruption issue
Date: Fri, 4 Apr 2014 18:40:35 +0100
User-agent: Mutt/1.5.21 (2010-09-15)

* Eric Blake (address@hidden) wrote:
> On 04/04/2014 03:57 AM, address@hidden wrote:
> > From: ChenLiang <address@hidden>
> > 
> > V5-->V4
> > * Fix two issues: one is cache_insert don't update the page which
> >   has been in the cache. Another avoiding the risk that run
> >   xbzrle_encode_buffer on changing data.
> 
> Is all or part of this series essential to have in 2.0 to avoid having a
> regression?

The first one (Fix one XBZRLE corruption issue) is probably worth it;
however my understanding is that it can only cause this corruption 
in the case where the source qemu runs out of RAM, and a few months
ago that used to be the case where the source qemu would abort at that
point, so it wasn't pleasent then.

Everything is else is an improvement.

Dave

> > 
> > 
> > a. Optimization the xbzrle remarkable decrease the cache misses.
> >     The efficiency of compress increases more than fifty times.
> >     Before the patch set, the cache almost totally miss when the
> >     number of cache item less than the dirty page number. Now the
> >     hot pages in the cache will not be replaced by other pages.
> > 
> > b. Reducing the data copy
> > 
> > c. Fix one corruption issues.
> > 
> > ChenLiang (10):
> >   XBZRLE: Fix one XBZRLE corruption issues
> 
> Based on name, this patch is worth including in the release, if it is
> not too late.
> 
> >   migration: Add counts of updating the dirty bitmap
> >   migration: expose the bitmap_sync_count to the end user
> >   migration: expose xbzrle cache miss rate
> 
> whereas these names sound like new features, and thus should wait for 2.1.
> 
> >   XBZRLE: optimize XBZRLE to decrease the cache misses
> >   XBZRLE: rebuild the cache_is_cached function
> >   xbzrle: don't check the value in the vm ram repeatedly
> >   xbzrle: check 8 bytes at a time after an concurrency scene
> >   migration: optimize xbzrle by reducing data copy
> >   migration: clear the dead code
> > 
> >  arch_init.c                    |  74 +++++++++++++++++-------------
> >  docs/xbzrle.txt                |   8 ++++
> >  hmp.c                          |   4 ++
> >  include/migration/migration.h  |   2 +
> >  include/migration/page_cache.h |  10 ++--
> >  migration.c                    |   3 ++
> >  page_cache.c                   | 101 
> > +++++++++++------------------------------
> >  qapi-schema.json               |   9 +++-
> >  qmp-commands.hx                |  15 ++++--
> >  xbzrle.c                       |  48 ++++++++++++++------
> >  10 files changed, 144 insertions(+), 130 deletions(-)
> > 
> 
> -- 
> Eric Blake   eblake redhat com    +1-919-301-3266
> Libvirt virtualization library http://libvirt.org
> 


--
Dr. David Alan Gilbert / address@hidden / Manchester, UK



reply via email to

[Prev in Thread] Current Thread [Next in Thread]