lkml.org 
[lkml]   [2005]   [Dec]   [30]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
    /
    Date
    From
    SubjectRe: [PATCH 6/9] clockpro-clockpro.patch
    Hi Rik!

    On Fri, Dec 30, 2005 at 08:22:12PM -0500, Rik van Riel wrote:
    > On Fri, 30 Dec 2005, Marcelo Tosatti wrote:
    >
    > > I think that final objective should be to abstract it away completly,
    > > making it possible to select between different policies, allowing
    > > further experimentation and implementations such as energy efficient
    > > algorithms.
    >
    > I'm not convinced. That might just make vmscan.c harder to read ;)

    Are you serious or just joking? :)

    Sure it might make it harder to read, but allowing selectable policies
    is very interesting. Peter's patches go half-way into that direction.

    Lets say, if CLOCK-Pro underperforms for a given workload (take into
    account that its simply optimizing reclaim for a subset of all existing
    access patterns, ie. heuristics), it would be easier for people to
    develop/use different policies.

    > > About CLOCK-Pro itself, I think that a small document with a short
    > > introduction would be very useful...
    >
    > http://linux-mm.org/AdvancedPageReplacement

    I meant something more like Documentation/vm/clockpro.txt, for easier
    reading of patch reviewers and community in general.

    > > > The HandCold rotation is driven by page reclaim needs. HandCold in turn
    > > > drives HandHot, for every page HandCold promotes to hot HandHot needs to
    > > > degrade one hot page to cold.
    > >
    > > Why do you use only two clock hands and not three (HandHot, HandCold and
    > > HandTest) as in the original paper?
    >
    > Because the non-resident pages cannot be in the clock.
    > This is both because of space overhead, and because the
    > non-resident list cannot be per zone.

    I see - that is a fundamental change from the original CLOCK-Pro
    algorithm, right?

    Do you have a clear idea about the consequences of not having
    non-resident pages in the clock?

    > I agree though, Peter's patch could use a lot more
    > documentation.

    -
    To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
    the body of a message to majordomo@vger.kernel.org
    More majordomo info at http://vger.kernel.org/majordomo-info.html
    Please read the FAQ at http://www.tux.org/lkml/

    \
     
     \ /
      Last update: 2005-12-31 04:33    [W:0.027 / U:0.208 seconds]
    ©2003-2016 Jasper Spaans. hosted at Digital OceanAdvertise on this site