[CentOS-virt] Recommendations, please

Thu Jun 23 16:19:39 UTC 2011
R P Herrold <herrold at owlriver.com>

On Thu, 23 Jun 2011, Tom Bishop wrote:

> Russ if you have time can you elaborate more about why you are continuing to
> go down the Xen path, I for one would love to hear the why's and what for.
> I can understand the hardware requirements, and I know xen is generally
> going to be faster but my small requirements have decided to start moving
> things to KVM since that is the direction of the upstream...would welcome
> your opinions if you have time available...Thanks in advance :)

I wasn't hiding my reasoning -- part of my reasoning is soft 
and 'touchy, feely' but you asked ... ;)

>> As part of that work (related to KVM hardware minimum
>> requirements, compatability with certain local libvirt based
>> tools, and performance of KVM vs. xen), I and other techs have

1. support for an existing hardware delivery base
 	-- I find that the upstream is falsely assuming
 	everything fielded has (or should) hardware virt
 	bits enabled.  This does not match the
 	refresh lifecycles I observe at _my_ customers,
 	nor at our shop

2. we have a substantial investment in libvirt tools which are
 	unproven and unqualified as to KVM until we get our
 	hands on the official 'as issued' CentOS 6.

 	I've been VERY frustrated with the update API compatability
 	in the upstream's 5 product line in this regard, as
 	gratutitous changes, not documented, creep in, and
 	remain unresolved for months, if not full point update
 	cycles.  I'm not a 'big enough fish' for those
 	customers running the 'supported' product to get
 	updates released, and so we are very conservative
 	in what we deploy

 	# 649438 still NEW, opened 2010-11-03, confirmed in
 		5.6 on 2011-05-26
 	# 506688 closed unfixed 2010-11-03, opened 2009-06-18

3. performance of KVM vs. xen -- formal metrics will be issued
 	by me once CentOS 6 issues, but in identical hardware
 	side by side tests, xen zips, and kvm waddles

 	Upstream has its investment in the KVM technology to
 	justify, and it may well be that things 'get better',
 	but if one is not pulling metrics against the
 	competition, one is deluding onself; I do not see
 	evidence that this is occurring

I put numbers on out production bottlenecks; we use agile 
techniques to address the 'hottest' issues daily, tdd to 
prevent unspecified behaviours from creeping in, and 'belt and 
suspender' techniques to design out recurrence errors in our 
processes.  KVM is 'not there yet' for me

-- Russ herrold