[RFC][PATCH] KVM: APPLES can improve the performance of applications and virtualized systems by up to 49%

[RFC][PATCH] KVM: APPLES can improve the performance of applications and virtualized systems by up to 49%

[Weiwei Jia]

Dear all,

Recently, we made a few improvements on effectively utilizing Pause
Loop Exiting (PLE) support for higher throughput on virtualized
systems. Basically, it solves two problems: 1) how to adjust
PLE_Window; 2) how to select virtual CPUs to schedule on VM_EXITs
caused by PLE. Our tests with standard benchmarks show that the
approach can improve performance by up to 49%. The approach shows
promising performance and is easy to implement. We think that it would
be wonderful if Linux/KVM and XEN can consider the approach.

We already have a prototype implementation based on KVM (Linux Kernel
3.19.8). Our patch for Linux Kernel 3.19.8 and the paper describing
our idea are available in Github repository [1][2][3]. We are pleased
to revise our patch in order to merge it into Linux/KVM and XEN. We
hope that you can test and adopt our approach/techniques. We are
pleased to get some comments/suggestions on the approach and on how
the idea can be adopted/tested by Linux/KVM and XEN. Thank you.

[1] APPLES paper: https://github.com/sysmen/apples/tree/master/paper
[2] APPLES patch:
https://github.com/sysmen/apples/blob/master/patches/3.19.8-APPLES.patch
[3] APPLES patch README:
https://github.com/sysmen/apples/blob/master/patches/README.txt

Best Regards,
Sysmen Research Group

Re: [Xen-devel] [RFC][PATCH] KVM: APPLES can improve the performance of applications and virtualized systems by up to 49%

[Wei Liu]

On Sat, May 12, 2018 at 04:27:04PM +0800, Weiwei Jia wrote:
> Dear all,
> 
> Recently, we made a few improvements on effectively utilizing Pause
> Loop Exiting (PLE) support for higher throughput on virtualized
> systems. Basically, it solves two problems: 1) how to adjust
> PLE_Window; 2) how to select virtual CPUs to schedule on VM_EXITs
> caused by PLE. Our tests with standard benchmarks show that the
> approach can improve performance by up to 49%. The approach shows
> promising performance and is easy to implement. We think that it would
> be wonderful if Linux/KVM and XEN can consider the approach.
> 
> We already have a prototype implementation based on KVM (Linux Kernel
> 3.19.8). Our patch for Linux Kernel 3.19.8 and the paper describing
> our idea are available in Github repository [1][2][3]. We are pleased
> to revise our patch in order to merge it into Linux/KVM and XEN. We
> hope that you can test and adopt our approach/techniques. We are
> pleased to get some comments/suggestions on the approach and on how
> the idea can be adopted/tested by Linux/KVM and XEN. Thank you.
> 
> [1] APPLES paper: https://github.com/sysmen/apples/tree/master/paper
> [2] APPLES patch:
> https://github.com/sysmen/apples/blob/master/patches/3.19.8-APPLES.patch
> [3] APPLES patch README:
> https://github.com/sysmen/apples/blob/master/patches/README.txt
> 

Is PV spinlock involved in your test? There is no mention of it in your
paper.

Wei.

> Best Regards,
> Sysmen Research Group
> 
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xenproject.org
> https://lists.xenproject.org/mailman/listinfo/xen-devel

Re: [RFC][PATCH] KVM: APPLES can improve the performance of applications and virtualized systems by up to 49%

[David Hildenbrand]

On 12.05.2018 10:27, Weiwei Jia wrote:
> Dear all,
> 
> Recently, we made a few improvements on effectively utilizing Pause
> Loop Exiting (PLE) support for higher throughput on virtualized
> systems. Basically, it solves two problems: 1) how to adjust
> PLE_Window; 2) how to select virtual CPUs to schedule on VM_EXITs
> caused by PLE. Our tests with standard benchmarks show that the
> approach can improve performance by up to 49%. The approach shows
> promising performance and is easy to implement. We think that it would
> be wonderful if Linux/KVM and XEN can consider the approach.
> 
> We already have a prototype implementation based on KVM (Linux Kernel
> 3.19.8). Our patch for Linux Kernel 3.19.8 and the paper describing
> our idea are available in Github repository [1][2][3]. We are pleased
> to revise our patch in order to merge it into Linux/KVM and XEN. We
> hope that you can test and adopt our approach/techniques. We are
> pleased to get some comments/suggestions on the approach and on how
> the idea can be adopted/tested by Linux/KVM and XEN. Thank you.
> 

Hi,

Please port the patch to latest upstream and send it as a proper patch
to this mailing list. (otherwise you won't get feedback on it WHP)

If you want some initial comment if this makes sense at all in the
context of KVM and can be implemented, you can send the patch itself as
RFC (based on an older kernel version).

Thanks!

> [1] APPLES paper: https://github.com/sysmen/apples/tree/master/paper
> [2] APPLES patch:
> https://github.com/sysmen/apples/blob/master/patches/3.19.8-APPLES.patch
> [3] APPLES patch README:
> https://github.com/sysmen/apples/blob/master/patches/README.txt
> 
> Best Regards,
> Sysmen Research Group
> 


-- 

Thanks,

David / dhildenb

Re: [Xen-devel] [RFC][PATCH] KVM: APPLES can improve the performance of applications and virtualized systems by up to 49%

[Dario Faggioli]

[-- Attachment #1: Type: text/plain, Size: 1126 bytes --]

On Sat, 2018-05-12 at 16:27 +0800, Weiwei Jia wrote:
> We already have a prototype implementation based on KVM (Linux Kernel
> 3.19.8). Our patch for Linux Kernel 3.19.8 and the paper describing
> our idea are available in Github repository [1][2][3]. We are pleased
> to revise our patch in order to merge it into Linux/KVM and XEN. We
> hope that you can test and adopt our approach/techniques. We are
> pleased to get some comments/suggestions on the approach and on how
> the idea can be adopted/tested by Linux/KVM and XEN. Thank you.
> 
To see how this is being handled in Xen currently, I suggest you to
grep for:
 ple_gap
 ple_window
 SECONDARY_EXEC_PAUSE_LOOP_EXITING
 EXIT_REASON_PAUSE_INSTRUCTION
 pauseloop_exits

I am aware of your paper, but I haven't got to being able to read it
carefully yet. I'll do that (but can't promise when).

Regards,
Dario
-- 
<<This happens because I choose it to happen!>> (Raistlin Majere)
-----------------------------------------------------------------
Dario Faggioli, Ph.D, http://about.me/dario.faggioli
Software Engineer @ SUSE https://www.suse.com/

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 833 bytes --]