[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <8ffd0aab-1adb-17a9-3055-ad60c31f8eb6@gmail.com>
Date: Fri, 17 Jun 2016 08:28:14 -0400
From: "Austin S. Hemmelgarn" <ahferroin7@...il.com>
To: Vitaly Wool <vitalywool@...il.com>,
Minchan Kim <minchan@...nel.org>
Cc: Geliang Tang <geliangtang@...il.com>,
Nitin Gupta <ngupta@...are.org>,
Sergey Senozhatsky <sergey.senozhatsky.work@...il.com>,
Dan Streetman <ddstreet@...e.org>,
LKML <linux-kernel@...r.kernel.org>,
Linux-MM <linux-mm@...ck.org>
Subject: Re: [PATCH] zram: update zram to use zpool
On 2016-06-17 04:30, Vitaly Wool wrote:
> Hi Minchan,
>
> On Thu, Jun 16, 2016 at 1:17 AM, Minchan Kim <minchan@...nel.org> wrote:
>> On Wed, Jun 15, 2016 at 10:42:07PM +0800, Geliang Tang wrote:
>>> Change zram to use the zpool api instead of directly using zsmalloc.
>>> The zpool api doesn't have zs_compact() and zs_pool_stats() functions.
>>> I did the following two things to fix it.
>>> 1) I replace zs_compact() with zpool_shrink(), use zpool_shrink() to
>>> call zs_compact() in zsmalloc.
>>> 2) The 'pages_compacted' attribute is showed in zram by calling
>>> zs_pool_stats(). So in order not to call zs_pool_state() I move the
>>> attribute to zsmalloc.
>>>
>>> Signed-off-by: Geliang Tang <geliangtang@...il.com>
>>
>> NACK.
>>
>> I already explained why.
>> http://lkml.kernel.org/r/20160609013411.GA29779@bbox
>
> This is a fair statement, to a certain extent. I'll let Geliang speak
> for himself but I am personally interested in this zram extension
> because I want it to work on MMU-less systems. zsmalloc can not handle
> that, so I want to be able to use zram over z3fold.
I concur with this.
It's also worth pointing out that people can and do use zram for things
other than swap, so the assumption that zswap is a viable alternative is
not universally correct. In my case for example, I use it on a VM host
for temporary storage for transient SSI VM's. Making it more
deterministic would be seriously helpful in this case, as it would mean
I can more precisely provision resources on this particular system, and
could better account for latencies in the testing these transient VM's
are used for.
Powered by blists - more mailing lists