Re: stochastic fair queueing in the elevator [Re: [BENCHMARK] 2.4.20-ck3 / aa / rmap with contest]

[Jakob Oestergaard <jakob@unthought.net>]

On Mon, Feb 10, 2003 at 03:58:26PM +1100, Nick Piggin wrote:
...
> >If we assume they are synchronous, that would be rather unfair
> >especially on multi-user systems - and the 90% accuracy that Rik
> >suggested would seem exaggerated to say the least (accuracy would be
> >more like 10% on a good day).
> >
> Remember that readahead gets scaled down quickly if it isn't
> getting hits. It is also likely to be sequential and in the
> track buffer, so it is a small cost.

I buy the sequential-argument - very good point.

Thanks,

> Huge readahead is a problem however anticipatory scheduling
> will hopefully allow good throughput for multiple read streams
> without requiring much readahead.

I'm curious as to how these things will interact in the NFS
server<->client situation  :)

Time will show, I guess.

Random data-point (not meant as a rant - I'm happy for all I got  ;)

In stock 2.4.20 the interaction is horrible - whatever was done there is
not optimal.    A 'tar xf' on the client will neither load the network
nor the server - it seems to be network latency bound (readahead not
doing it's job - changing min-readahead and max-readahead on the client
doesn't seem to make a difference). However, my desktop (running on the
client) can hang for 10 seconds straight every half hour or so, when the
nightly backup runs on the server (local disk reads streaming at around
2 MB/sec from an array capable of at least 40 MB/sec).

-- 
................................................................
:   jakob@unthought.net   : And I see the elder races,         :
:.........................: putrid forms of man                :
:   Jakob Østergaard      : See him rise and claim the earth,  :
:        OZ9ABN           : his downfall is at hand.           :
:.........................:............{Konkhra}...............: