Re: Some Xen performance measurements

[David TAILLANDIER <david.taillandier%domainename.com@localhost>]

Martti Kuparinen <martti.kuparinen%iki.fi@localhost> wrote:
> Remarks and questions
> =====================
>
> - One would assume that running 4 VMs would take 4x the compilation
> time compared to 1 VM. Any ideas why the time is only 1600 instead of
> 2000?


Some people argued you should use make -j4 but the experiment lead to 
the same results. IMHO this is because the CPU are alredy nearly at 100%.

My first explanation was : you launched all 'make' simultaneously, then 
the internal cache of the hard-drive shows its power.

I then tested as follow :

cpu0 at mainbus0: (uniprocessor)
cpu0: AMD Athlon XP 2000+ (686-class), 1674.51 MHz, id 0x662

To eliminate the hard-drive cache, I launch the tasks manually with some 
delay. And to obtain something correct, I launch the task tree times. I 
then only take the results from the 'central' makes.

time --------------------------------------------------------->
VM 1 : [----make----][----make----][----make----]
VM 2 :   [----make----][----make----][----make----]
VM 3 :     [----make----][----make----][----make----]
VM 4 :       [----make----][----make----][----make----]


1 x NetBSD 128 MB XENU
----------------------
695.71 real       582.11 user        86.34 sys

4 x NetBSD 128 MB XENU
----------------------
2938.47 real      2196.85 user       354.63 sys
2898.55 real      2211.86 user       323.55 sys
2944.94 real      2379.09 user       366.46 sys
2925.15 real      2233.20 user       312.56 sys

We then see that 4 x NetBSD perform a little slower than one x 4.
I did not test anything else, this is too long for me  :-)