lkml.org 
[lkml]   [1998]   [Sep]   [22]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: Today Linus redesigns the networking driver interface (was Re: tulip driver in ...)
Linus Torvalds wrote:

> That's essentially the rule now. It's simple, and it works. However, it
> doesn't give us aggregation - which could improve throughput and CPU
> usage.
>
> (Aggregation tends to be bad for latency, which is why it's hard: there
> needs to be some way for the interrupt handler to tell the system that "I
> want bh's run _now_, because I got an important packet or I'm close to
> filling up the queues").
>
> I don't know if it's really worth pursuing, though.

Doubtful. From the SCSI standpoint, aggregation is very bad. At least in
the 2.0.x kernel series, I once made my aic7xxx driver store up completions
until either the tq_scheduler was ran, or as a backup when tq_timer was
ran. Whenever the commands where completed on tq_scheduler (aka, at
interrupt exit and without aggregation any more than the single interrupt
offered) then performance was just like normal. When the commands that were
completing happened to by synchronous swaps, then the tq_scheduler didn't
get ran at the end of each interrupt, so we had to wait for tq_timer to
catch us, which limited our swap throughput to 100 ops/sec for each swap
device. It sucked so bad that it was sufficient reason enough to scrap the
whole setup.

--

Doug Ledford <dledford@dialnet.net>
Opinions expressed are my own, but
they should be everybody's.

-
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 13:44    [W:0.143 / U:0.244 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site