Re: Problems with many DOMUs on a single DOM0.

[Harry Waddell <waddell%caravaninfotech.com@localhost>]

On Tue, 8 Jan 2013 06:11:19 +0530
"Cherry G. Mathew" <cherry.g.mathew%gmail.com@localhost> wrote:

> On 8 January 2013 05:33, Harry Waddell 
> <waddell%caravaninfotech.com@localhost> wrote:
> 
> > * I ran some benchmarks, although I didn't collect enough samples to make 
> > it scientific,
> > using bonnie++ -s 8g, and using lvm backed ext4 filesystems were faster for 
> > everything except rewriting data. Sequential blocks in and out were 
> > noticably better for the lvm backed case. I'll be happy to share my notes 
> > if people are interested.
> 
> Any numbers you have would always be interesting - especially in
> comparison to other things ( in this case file-backed VMs, or linux
> dom0 for eg:).
> 
> Cheers,
> 
> -- 
> ~Cherry
> 
> 

ok, here you go. I mostly had linux numbers, so I ran a few tests to add some 
netbsd ones to the set, including some dom0 numbers. ffs filesystems are all 
mounted with -o log and made using newfs -O2. Linux filesytems use default mkfs 
options and mount defaults. 

http://www.extremelymodern.com/xen-fs-performance.pdf

ideally, I'd run each of these tests several times and create an average, but I 
think the numbers are at least decent approximations of the truth.

All tests were run on the same hardware, 2 x E2630 cpu, 4 x WD ent drives, RAID 
5 using a 3ware 9650-4LP controller, all non-testing systems as idle as 
possible. 

On the plus side, my tests show that ffsv2 works quite well, even when compared 
with some of the better filesystems that linux has to offer. 


Harry Waddell