| |||||||||||||
TypoTypoPosted Dec 19, 2010 4:47 UTC (Sun) by quotemstr (subscriber, #45331)In reply to: Typo by Halmonster Parent article: Unmapped page cache control
Doesn't having the host use uncached IO to service virtualized guest IO eliminate the duplicate caching?
(Log in to post comments)
Obvious Answer? Posted Dec 19, 2010 6:42 UTC (Sun) by Fowl (subscriber, #65667) [Link]
Or just the disabling the cache on the guests? Thereby allowing them to all share the host's.
What am I missing?
Obvious Answer? Posted Dec 19, 2010 7:07 UTC (Sun) by quotemstr (subscriber, #45331) [Link]
It's easier and less invasive to configure a program to use unbuffered IO for a particular program's access to a particular file (a virtual disk in this case) than to disable buffering for a whole operating system, which might be actually be impossible.
Typo Posted Dec 20, 2010 11:39 UTC (Mon) by joern (subscriber, #22392) [Link]
> Doesn't having the host use uncached IO to service virtualized guest IO eliminate the duplicate caching?
It does - if only a single guest is caching the pages in question.
Interesting Posted Dec 20, 2010 14:32 UTC (Mon) by rilder (subscriber, #59804) [Link]
I don't think that is possible -- to be able to differentiate requests from the host and the guest to serve uncached to guest and cached to host -- if this is possible then it is great.
Interesting Posted Dec 20, 2010 14:39 UTC (Mon) by quotemstr (subscriber, #45331) [Link]
It's not only possible, but trivial: IO from guests goes through the hypervisor, which can use simple uncached file IO (e.g., O_DIRECT) to service these requests. Host IO goes through the normal path and uses normal buffering.
Interesting Posted Jan 3, 2011 5:17 UTC (Mon) by balbir_singh (subscriber, #34142) [Link]
Yes, it is possible, but there have been reports of large overheads in doing so. Please see the email from Chris at http://www.mail-archive.com/kvm@vger.kernel.org/msg30821.... for throughput issues.
|
|||||||||||||