towards an accurate delay....
Andrew P. Black
black at cse.ogi.edu
Tue Mar 12 07:15:10 UTC 2002
Some of the systems folk here at OGI have been experimenting with
something that we call "firm timers" in the Linux kernel. We can get
microsecond timers with low overhead.
The idea is similar to the Soft Timers [1] work from the last SOSP,
but with a hardware interrupting device as a backup. Mostly, the
hardware device is turned off before it triggers, so there is no
interrupt processing overhead.
The idea of soft timers is that the system checks a clock register
whenever it gets control (e.g., because of someone else's system
call), and wakes up your process at the right time.
How does any of this apply to an application rather than the kernel?
Well, the application can't get a lower latency than the kernel is
able to provide, so anything that we do inside Squeak would seem to
be limited by what the kernel is offering to the VM. And, if the
kernel is offering high-resolution, low cost timers, the VM could
just use them.
Andrew
[1] Mohit Aron and Peter Drushel. SOft Timers: efficient microsecond
software timer support for network processing. In ACM TOCS 18(3)
August 2000.
More information about the Squeak-dev
mailing list
|