[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <c8ea7076-f107-c340-6b34-6b22a9369da5@redhat.com>
Date: Fri, 23 Dec 2022 10:40:17 +0100
From: David Hildenbrand <david@...hat.com>
To: Soichiro Ueda <the.latticeheart@...il.com>, mst@...hat.com,
jasowang@...hat.com, akpm@...ux-foundation.org
Cc: virtualization@...ts.linux-foundation.org,
linux-kernel@...r.kernel.org, linux-mm@...ck.org,
kalutes@...gle.com, mhiramat@...nel.org, cwd@...gle.com
Subject: Re: [PATCH] virtio_balloon: high order allocation
On 23.12.22 10:35, Soichiro Ueda wrote:
> At present, the VirtIO balloon device driver allocates pages
> one by one using alloc_page(), and frees them using put_page().
>
> This changes it so that the driver allocates high order pages
> by using alloc_pages(), and frees them using __free_pages() if possible.
> By doing so, the CPU performance of inflation and deflation
> should be improved.
>
> The effect of this change has been confirmed by benchmarks that measure
> the elapsed time of inflation and deflation.
>
> The results are here:
>
> 16 pages inflation:
> before: 119,779 ns
> after : 115,655 ns (-3.4%)
> 64 pages inflation:
> before: 156,977 ns
> after : 150,961 ns (-3.8%)
> 256 pages inflation:
> before: 218,649 ns
> after : 208,490 ns (-4.6%)
> 16 pages deflation:
> before: 78,112 ns
> after : 68,288 ns (-12.6%)
> 64 pages deflation:
> before: 97,205 ns
> after : 80,491 ns (-17.194%)
> 256 pages deflation:
> before: 122,742 ns
> after : 107,526 ns (-12.4%)
How does this affect page migration / balloon compaction etc?
--
Thanks,
David / dhildenb
Powered by blists - more mailing lists