lkml.org 
[lkml]   [2000]   [Mar]   [9]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: elevator messages in 2.3.50
Andrea Arcangeli wrote:
> >I don't know how you reach this conclusion. Can you explain?
> >
> >I can see that if you never receive requests during holdoff then of
> >course you lose. [..]
>
> You answered your question yourself.
>
> Consider the worst case of the heuristic. Its worts case is definitely the
> common simple case. Harming it is a no way IMHO.

That's what I don't see. That the worst case is the _common_ simple
case. Sure it's the simpler case, but I see random paging as a common
case which is the best case for this heuristic.

You _might_ be right, but do you know for sure? I.e. has it ever been
measured?

Two more things:

- Using numbers plucked from the air, the heuristic's worst case will
reduce throughput by at most 2%, and the best case can improve
throughput by several times. So the "typical average behaviour" may
improve even if the heuristic doesn't reorder requests most of the
time.

- You can make it self-tuning w.r.t. whether it is worth doing
according to recent access patterns. To do this, if a request
arrives within 100us if a long seek request being issued, and the
request which arrived would have been a short seek, then the
heuristic would have been a gain. It may be that monitoring the
occurrence of that indicates the future likelihood of the heuristic
being a gain. (Again, need's measurement).

In this way even the worst case you are thinking of is avoided.

-- Jamie


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

\
 
 \ /
  Last update: 2005-03-22 13:56    [W:0.035 / U:0.048 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site