Re: [Qemu-devel] [PATCH] implement moderation registers for e1000

[Stefan Hajnoczi]

On Fri, Feb 8, 2013 at 11:46 AM, Paolo Bonzini <address@hidden> wrote:
> Il 08/02/2013 11:20, Luigi Rizzo ha scritto:
>> On Fri, Feb 8, 2013 at 2:02 AM, Stefan Hajnoczi <address@hidden
>> <mailto:address@hidden>> wrote:
>>
>>     On Wed, Feb 06, 2013 at 03:23:41PM +0100, Luigi Rizzo wrote:
>>     > The following patch implements interrupt moderation registers
>>     > for the e1000 adapter. These feature is normally used by OS
>>     > drivers, and their implementation improves performance significantly,
>>     > especially on the transmit path.
>>     > The feature can be disabled through a command line option.
>>     > We have seen only benefits in throughput, while latency slightly
>>     > increases (that is an inherent feature of interrupt moderation)
>>     > because very short delays cannot be emulated precisely.
>>     >
>>     > For those curious on performance, there is a set of measurements
>>     > (of this and other changes that we will post separately) at
>>     >
>>     > http://info.iet.unipi.it/~luigi/research.html#qemu
>>
>>     http://info.iet.unipi.it/~luigi/papers/20130206-qemu.pdf is 404.
>>
>>
>> sorry, fixed now.
>> And, will resubmit a fixed patch with uninit and fixed braces in the
>> if() statement.
>>
>> I am happy to make this default to off. But it would be good if you could
>> actually give it a try. Note that linux and FreeBSD (and presumably windows)
>> do use moderation by default so enabling the emulation of the
>> registers makes the guest OS behave differently (and closer to bare metal).
>>
>> To test that the patch itself does not cause regression in the emulator
>> one should also turn off moderation in the guest (one of the tests i
>> have run).
>
> I think making the default on is fine, but you need to add compatibility
> options to leave it off in older machine types (pc-1.4 and earlier).

Latency regression.  Would need to see real results to understand how bad it is.

Stefan