Date:   Mon, 3 Jun 2019 11:31:51 +0200
From:   David Hildenbrand <david@...hat.com>
To:     Alexander Duyck <alexander.duyck@...il.com>, nitesh@...hat.com,
        kvm@...r.kernel.org, mst@...hat.com, dave.hansen@...el.com,
        linux-kernel@...r.kernel.org, linux-mm@...ck.org
Cc:     yang.zhang.wz@...il.com, pagupta@...hat.com, riel@...riel.com,
        konrad.wilk@...cle.com, lcapitulino@...hat.com,
        wei.w.wang@...el.com, aarcange@...hat.com, pbonzini@...hat.com,
        dan.j.williams@...el.com, alexander.h.duyck@...ux.intel.com
Subject: Re: [RFC PATCH 00/11] mm / virtio: Provide support for paravirtual
 waste page treatment

On 30.05.19 23:53, Alexander Duyck wrote:
> This series provides an asynchronous means of hinting to a hypervisor
> that a guest page is no longer in use and can have the data associated
> with it dropped. To do this I have implemented functionality that allows
> for what I am referring to as "waste page treatment".
> 
> I have based many of the terms and functionality off of waste water
> treatment, the idea for the similarity occured to me after I had reached
> the point of referring to the hints as "bubbles", as the hints used the
> same approach as the balloon functionality but would disappear if they
> were touched, as a result I started to think of the virtio device as an
> aerator. The general idea with all of this is that the guest should be
> treating the unused pages so that when they end up heading "downstream"
> to either another guest, or back at the host they will not need to be
> written to swap.
> 
> So for a bit of background for the treatment process, it is based on a
> sequencing batch reactor (SBR)[1]. The treatment process itself has five
> stages. The first stage is the fill, with this we take the raw pages and
> add them to the reactor. The second stage is react, in this stage we hand
> the pages off to the Virtio Balloon driver to have hints attached to them
> and for those hints to be sent to the hypervisor. The third stage is
> settle, in this stage we are waiting for the hypervisor to process the
> pages, and we should receive an interrupt when it is completed. The fourth
> stage is to decant, or drain the reactor of pages. Finally we have the
> idle stage which we will go into if the reference count for the reactor
> gets down to 0 after a drain, or if a fill operation fails to obtain any
> pages and the reference count has hit 0. Otherwise we return to the first
> state and start the cycle over again.

While I like this analogy, I don't like the terminology mixed into
linux-mm core.

mm/aeration.c? Bubble? Treated? whut?

Can you come up with a terminology once can understand without a PHD in
biology? (if that is even the right field of study, I have no idea)


ALSO: isn't the analogy partially wrong? Nobody would be using "waste
water" just because they are low on "clean water". At least not in my
city (I hope so ;) ). But maybe I am not getting the whole concept
because we are dealing with pages we want to hint to the hypervisor and
not with actual "waste".

> 
> This patch set is still far more intrusive then I would really like for
> what it has to do. Currently I am splitting the nr_free_pages into two
> values and having to add a pointer and an index to track where we area in
> the treatment process for a given free_area. I'm also not sure I have
> covered all possible corner cases where pages can get into the free_area
> or move from one migratetype to another.

Yes, it is quite intrusive. Maybe we can minimize the impact/error
proneness.

> 
> Also I am still leaving a number of things hard-coded such as limiting the
> lowest order processed to PAGEBLOCK_ORDER, and have left it up to the
> guest to determine what size of reactor it wants to allocate to process
> the hints.
> 
> Another consideration I am still debating is if I really want to process
> the aerator_cycle() function in interrupt context or if I should have it
> running in a thread somewhere else.

Did you get to test/benchmark the difference?

-- 

Thanks,

David / dhildenb

Powered by blists - more mailing lists