hi Daniel,
I think this is the paper that Adam Huffman mentioned in passing
during GridPP33, in fact.
(It's certainly something that's still a problem with VMs and needs
taken seriously.)
Sam
On 26 August 2014 10:39, Daniel Traynor <[log in to unmask]> wrote:
> Given some of the discussion we had at gridpp33 on clouds and VM performance I thought this paper from IBM was timely
>
> http://domino.research.ibm.com/library/cyberdig.nsf/papers/0929052195DD819C85257D2300681E7B/$File/rc25482.pdf
>
> from the summary
>
> "KVM still adds some overhead to every I/O
> operation; this overhead ranges from significant when per-
> forming small I/Os to negligible when it is amortized over
> large I/Os. Thus, KVM is less suitable for workloads that
> are latency-sensitive or have high I/O rates. These overheads
> significantly impact the server applications we tested."
>
> Which might be relevant for analysis type workloads.
>
> dan
>
> * Dr Daniel Traynor, Grid cluster system manager
> * Tel +44(0)20 7882 6560, Particle Physics,QMUL
|