Re: [Xen-users] New binary release of GPL PV drivers for Windows

[Pasi Kärkkäinen <pasik@xxxxxx>]

On Thu, Feb 21, 2008 at 09:08:32PM +0100, Stephan Seitz wrote:
> Hi Pasi,
> 
> I wasn't able to get Windows XP Professional x64 running with gplpv until
> James released 0.8.0 if his great drivers.
> 
> So my answer is a bit delayed:
> 
> Equipment: core2 duo, 2,66GHz, Areca PCI-X Raid 6 over 8 disks
> 
> 
> System is running Xen 3.2.0 64bit, dom0 is 2.6.18.8
> 
> Tested HVM domU is running XP Pro x64, Version 2003, SP 2, tested with 
> iometer 2006-07-27 stable.
>

Thanks for the testing results!

So it seems windows gplpv drivers give almost 4x better performance compared to 
emulated
qemu disks.. Nice!

-- Pasi
 
> --- Disk/IO using a gplpv'd disk:
> 
> pattern: 4k, 50% read, 50% write
> 
>       total iops:             ~14180
>               read            ~7045-7065
>               write           ~7025-7045
>       total MB/s:             ~55
>               read            ~27.5
>               write           ~27.5 (looks like 50%...)
> 
>       avg IO response time:   ~0.071 ms
>       max IO response time:   ~19.438 ms
>       cpu utilization:        0% (??)
> 
> pattern: 32k, 50% read, 50% write
> 
>       total iops:             ~6900
>               read            ~3435
>               write           ~3450
>       total MB/s:             ~215
>               read            ~107.5
>               write           ~107.5
> 
>       avg IO response time:   ~0.145 ms
>       max IO response time:   ~21.525 ms
>       cpu utilization:        ~5.52%
> 
> 
> pure read operations with 32k pattern shows about 280 MB/s throughput
> pure write operation with 512B pattern shows about 8.5 MB/s througput
> 
> 
> --- Disk/IO using a QEMU disk:
> 
> pattern: 4k, 50% read, 50% write
> 
>       total iops:             ~3650
>               read            ~1828
>               write           ~1790
>       total MB/s:             ~14
>               read            ~7
>               write           ~7
> 
>       avg IO response time:   ~0.276 ms
>       max IO response time:   ~55.242 ms
>       cpu utilization:        98.7%
> 
> pattern: 32k, 50% read, 50% write
> 
>       total iops:             ~3064
>               read            ~1370-1390
>               write           ~1360
>       total MB/s:             ~84
>               read            ~42-44
>               write           ~40-42
> 
>       avg IO response time:   ~0.387 ms
>       max IO response time:   ~77.450 ms
>       cpu utilization:        ~76.8%
> 
> 
> pure read operations with 32k pattern shows about 94 MB/s throughput
> pure write operation with 512B pattern shows about 1.8 MB/s througput
> 
> 
> 
> --- (filebased) disk IO at dom0 (random, using dbench):
> 
>       10 workers on ext3,defaults: ~660 MB/s
>       10 workers on xfs,defaults: ~620 MB/s
> 
> hdparm shows 3.3 GB/s cached and 366 MB/s buffered
> 
> 
> 
> Pasi Kärkkäinen schrieb:
> >On Tue, Feb 05, 2008 at 10:22:55AM +0200, Pasi Kärkkäinen wrote:
> >>On Mon, Feb 04, 2008 at 06:21:41PM +0200, Pasi Kärkkäinen wrote:
> >>>On Sun, Feb 03, 2008 at 12:30:51PM +0100, Stephan Seitz wrote:
> >>>>If someone knows a Windows Benchmarking Suite, I'll do real tests. I 
> >>>>know my
> >>>>recent tests are not comparable to any value, but I'm a little bit 
> >>>>handycapped
> >>>>on windows ;)
> >>>>
> >>>You could use IOmeter http://www.iometer.org. It's a widely used disk
> >>>benchmarking tool on Windows.
> >>>
> >>>It's easy to run benchmarks using different requests sizes, different 
> >>>number
> >>>of outstanding io's etc..
> >>>
> >>With small requests sizes (512 bytes or 4k) you can measure how many IOPS
> >>(IO operations per second) you can get, and with big request sizes (64+ 
> >>kB)
> >>you can measure how much throughput you can get..
> >>
> >>Changing the number of outstanding IO's means how many IO operations are
> >>active at the same time (optimal values depends on the storage used, and 
> >>on
> >>the queue depth of the hardware, drivers and kernel). 
> >>
> >>Note that IOmeter wants to use raw disk devices, so don't create any
> >>partitions or format the disk before using IOmeter.
> >>
> >
> >And please share your benchmarking results :)
> >
> >-- Pasi
> 
> 

_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-users