Re: [PATCH][RFC] Supporting Enlightened Windows2008Server

Re: [PATCH][RFC] Supporting Enlightened Windows2008Server

[Ky Srinivasan]

>>> On Sat, Apr 5, 2008 at  5:21 AM, in message
<C41D032F.1600F%keir.fraser@eu.citrix.com>, Keir Fraser
<keir.fraser@eu.citrix.com> wrote: 
> On 5/4/08 00:24, "Ky Srinivasan" <ksrinivasan@novell.com> wrote:
> 
>> Based on the feedback I got from you and Tim, I am enclosing the next 
> version
>> of the patches to support enlightened win2008 server. Here are the changes I
>> have made:
>> 
>> 1) I have put the shim on a low calorie diet - I have gotten rid of the
>> framework infrastructure and to the extent possible integrated the shim code
>> with xen.
>> 
>> 2) I have tried to cleanup the code. I am sure  more work will be needed 
> here.
>> 
>> 3) I am not advertising the TLB related enlightenments. We can revisit this
>> later if needed.
> 
> It's certainly quite a bit shorter which is good. For the remaining stuff,
> do you have empirical evidence that performance is improved by it?
> 
I do. On the NetBench runs we did, we were seeing close to 10% improvement. I will keep you posted on other benchmark results as they become available. Furthermore, as MSFT defines new enlightenments, we can implement those cleanly within this body of code. Additionally, there is value in claiming that we support all the enlightenments needed for hosting enlightened win 2008 servers.

> Other more minor comments are that the coding style is still off (e.g.,
> start braces should go on their own line, spaces inside () for if/for/while
> headers), you have at least one big switch statement where most of the cases
> could be collapsed to just one shared block of code, and indeed shouldn't
> the 'default' case in the hypercall demuxing switch statement be to return
> 'denied', and that would get rid of most of the individual cases altogether?

In the next version of these patches, I will address the issues you have raised here.


Thanks,

K. Y
> 
>  -- Keir

Re: [PATCH][RFC] Supporting Enlightened Windows2008Server

[Keir Fraser]

On 7/4/08 15:28, "Ky Srinivasan" <ksrinivasan@novell.com> wrote:

>> It's certainly quite a bit shorter which is good. For the remaining stuff,
>> do you have empirical evidence that performance is improved by it?
>> 
> I do. On the NetBench runs we did, we were seeing close to 10% improvement. I
> will keep you posted on other benchmark results as they become available.
> Furthermore, as MSFT defines new enlightenments, we can implement those
> cleanly within this body of code. Additionally, there is value in claiming
> that we support all the enlightenments needed for hosting enlightened win 2008
> servers.

Do you know which hypercalls that benefit comes from? I see no reason to
support more hypercalls than necessary (I certainly don't see the value in
that as an end in itself).

 -- Keir

Re: [PATCH][RFC] Supporting Enlightened Windows2008Server

[Keir Fraser]

On 7/4/08 15:28, "Ky Srinivasan" <ksrinivasan@novell.com> wrote:

> I do. On the NetBench runs we did, we were seeing close to 10% improvement. I
> will keep you posted on other benchmark results as they become available.

I forgot to ask in my previous email what the configuration was for this
netbench test. For example, was it a UP or SMP guest? Where did the netbench
clients run? Did you run with PV drivers?

The latter is particularly relevant, as measuring I/O performance of an HVM
guest without PV I/O drivers is particularly useful. Unfortunately I'm not
sure whether anyone has got the GPL PV drivers working with Server 2008 as
yet. But if not that would seem a more useful step towards w2k8 performance
than Viridian emulation, imo. I bet if you compare XP performance on
netbench with versus without the GPL PV drivers, the difference is more than
10%!

 -- Keir

Re: [PATCH][RFC] Supporting Enlightened Windows2008Server

[Ky Srinivasan]

>>> On Mon, Apr 7, 2008 at 12:16 PM, in message
<C420076F.1EE11%keir.fraser@eu.citrix.com>, Keir Fraser
<keir.fraser@eu.citrix.com> wrote: 
> On 7/4/08 15:28, "Ky Srinivasan" <ksrinivasan@novell.com> wrote:
> 
>> I do. On the NetBench runs we did, we were seeing close to 10% improvement. 
> I
>> will keep you posted on other benchmark results as they become available.
> 
> I forgot to ask in my previous email what the configuration was for this
> netbench test. For example, was it a UP or SMP guest? Where did the netbench
> clients run? Did you run with PV drivers?

The server was a UP windows 2008 server guest. I suspect SMP numbers may actually show more of an improvement. The clients were run on physical machines. We used our PV drivers for windows (2008 server) to maximize the I/O performance.

> 
> The latter is particularly relevant, as measuring I/O performance of an HVM
> guest without PV I/O drivers is particularly useful. Unfortunately I'm not
> sure whether anyone has got the GPL PV drivers working with Server 2008 as
> yet. But if not that would seem a more useful step towards w2k8 performance
> than Viridian emulation, imo. I bet if you compare XP performance on
> netbench with versus without the GPL PV drivers, the difference is more than
> 10%!

Agreed; without the PV drivers, it does not make any sense to run NetBench. We have PV drivers that work with Windows 2008 server.

Regards,

K. Y
> 
>  --   Keir
> 
> 
> 
> _______________________________________________
> Xen-  devel mailing list
> Xen-  devel@lists.xensource.com
> http://lists.xensource.com/xen-  devel