[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

RE: [Xen-users] Release 0.9.5 of GPL PV Drivers for Windows


  • To: "Joost van den Broek" <jvandenbroek@xxxxxxxxx>
  • From: "James Harper" <james.harper@xxxxxxxxxxxxxxxx>
  • Date: Sat, 7 Jun 2008 21:37:37 +1000
  • Cc: xen-users@xxxxxxxxxxxxxxxxxxx
  • Delivery-date: Sat, 07 Jun 2008 04:38:14 -0700
  • List-id: Xen user discussion <xen-users.lists.xensource.com>
  • Thread-index: AcjIkRMvk/idNpbdSKCkj1nQtSKlGgAAQVQg
  • Thread-topic: [Xen-users] Release 0.9.5 of GPL PV Drivers for Windows

> 
> I'm sorry to react on a random posted message, I just subscribed this
> mailinglist. I've currently installed the 0.9.6 drivers which are a
> major step forward since the last time I tried your drivers (0.8.x).
The
> sustained transferrate of the scsi driver might not be as good as it
> could be, but on the other hand I am very impressed by it's I/O speed
> measured with IOMeter. It even outperformed the closed-source driver
> found in XenServer 4.0.1! Well that test was run on possibly slower
> disks, but same CPU, RAM and 4xRAID5 SATA.

Hmmmm... now I'm baffled. I'm getting reports of really high performance
with the newer releases, and other reports of really poor performance.

> Also "reallife" performance, as in just using the system, is at least
to
> call impressive. Very responsive, definitely enterprise ready. There
is
> just one thing I'm struggling with. Somehow the Tx network traffic
(seen
> from the HVM) is to DomU's and other nodes on the lan is very slow. On
> the other hand it's just fine to Dom0 (about 1Gbit/s). Below are some
> measurements done with iperf, all executed from the Windows HVM:
> 
> I tried enabling/disabling tx checksumming with ethtool on the HVM's
vif
> interface, changed the Xen Net Device Driver options, all with no
luck.

I've just been examining the code that controls that and it's not
working, so gso, sg, and csum offload are all 'stuck' in the on state,
regardless of what you set the device controls to. I'll investigate.

> These tests were performed on a AMD Opteron 1210 1.8ghz, HVM and
DomU's
> running 512MB RAM and Dom0 running on Debian Etch with Xen 3.2.0 from
> backports. The tested Windows HVM is running Windows Server 2003 R2
SP2
> EN 32-bit set to 1 vcpu with the latest GPL PV drivers (0.9.6)
> installed. The onboard NIC is a Broadcom BCM5721, not the greatest,
but
> that shouldn't matter in this case.

That sounds identical to my test environment. You wouldn't be running a
HP ML115 would you?

> The Linux PV's are performing well,
> so there has to be something wrong with the GPL PV drivers or
settings.
> Any suggestion?

Not that this point. All I've really tested with the network is DomU <->
Dom0 and, to a lesser extent, DomU <-> physical LAN.

I've just fixed a bug which was preventing formatting of a xenvbd device
under Windows and am just getting back to being able to use 'xm
block-detach' without getting a BSoD. If I can fix that quickly I'll do
a binary release tonight.

James

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


 


Rackspace

Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.