Messages in this thread | | | Date | Wed, 1 Dec 1999 00:35:20 +0100 (CET) | From | Ingo Molnar <> | Subject | Re: lowlatency-2.2.13-A1 questions |
| |
On Tue, 30 Nov 1999, William Montgomery wrote:
> The udelay_resched(n) macro does not seem to reference "n",
(yep, thats a bug)
> would the following be an acceptable implementation? > > #define udelay_resched(n) (\ > { int i; \ > for (i = 0; i < ((n+9)/10); i++) { \ > conditional_schedule(); \ > udelay(10); \ > } \ > })
yep, thats ok.
> I have been using the lowlatency-2.2.10-N6B.patch with good results. > I recently installed the 2.2.13 kernel and applied the lowlatency-2.2.13-A1 > patch, however the results were not quite as good. The 2.2.10-N6B patch > on the 2.2.10 kernel gave +/-500usec scheduling latencies whereas > the 2.2.13-A1 patch on the kernel gave +/-3msec latencies. Any idea > why? These results were obtained on a Pentium III 500MHz.
the 2.2.13 patches do not have the console changes - and text console scrollback is introducing latencies around 4msec. The reason is that occasionally the console driver has to read+copy the whole (text) videoram, and this is slow. I solved this by adding a read-cache for all videoram contents, thus videoram only ever has to be written. (which is fast)
> How were the high latency areas identified? I would be very happy > to help in testing and contributing to this patch.
i'll eventually forward port the console changes as well.
-- mingo
- 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/
| |