Messages in this thread | | | From | kuznet@ms2 ... | Subject | Re: Task queue makes slow device even slower | Date | Thu, 28 Oct 1999 18:27:31 +0400 (MSK DST) |
| |
Hello!
> As a long-term solution, perhaps something along the lines of the "soft > timers" described in http://www.cs.rice.edu/~aron/papers/soft-timers.ps.gz > might be useful for many cases where we have udelays or even mdelays > today.
Soft timers _are_ _implemented_ in linux from the very beginning and called BHs. Anyone who wants polling may hook them and poll. Essentially, packet scheduler in 2.2 uses soft timers, with events generated by net, system scheduler and timer.
I remember I've already said you and Jamal that I tried to add more events to soft timers in packet scheduler and the result was disguisting, they eat gobs of cpu and however inappropriate for real time tasks.
From the other hand, true variable resulution timer is possible and Ingo said he is ready to complete it.
What's about the paper. I read it not enough carefully, and, probably, it is the reason why I understood almost nothing. Seems, the experimental results summarized there are not quite legal. F.e. the thing puzzled me is increasing throughput on high delay*throuput pipes. It is the only place where timed TCP transmission have a sense, but they did no experiments emulating fat pipe and used normal one... Seems, in this case all the effect is achieved because disabling slow start. Well, seems, I have to read it more carefully.
Alexey Kuznetsov
- 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/
| |