From: Frank Geurts (geurts_at_rice.edu)
Date: Fri Mar 07 2003 - 16:19:29 EST
-------- Original Message --------
Subject: Re: FYI: L2 latencies
Date: Fri, 7 Mar 2003 16:08:21 -0500 (EST)
From: "Tonko A. Ljubicic" <tonko_at_bnl.gov>
To: Zoran Milosevich <zoran.milosevich_at_cmu.edu>
CC: Frank Geurts <geurts_at_rice.edu>, <startrig-hn_at_connery.star.bnl.gov>
On Fri, 7 Mar 2003, Zoran Milosevich wrote:
> Hi Frank -
> OK, thanks for the information. Anything before run 4065050, we can
perhaps
> understand. After this run, L2 was updated, so there should not be a
problem.
> At least from l2. Below is the Level 2 logging information we have
for that run.
> NOTES - l2Accept is the time between when l2 gets a token from l1 and
sends
> a L2_ACCEPT to l1. l2Process is the time between when sending an
L2_ACCEPT
> and sending token (and data) to GB. Units are in msec. I only flag
here times of
> L2Process greater than 5.0 OR l2Accept greater than 2.0. I see one
11 msec and
> a couple 7 msec L2_ACCEPT times, but nothing greater than that.
> I don't know if this helps you track this down.
[...]
I ran some tests just now and, sure enough, things are greatly improved.
The percoent of events above 10 ms is very small (0.1%, down from 10%) in
the first 10000 events. I certainly see more than 30-40 ms sporadically.
I actually did some measurements and I see an average time between a L0
issued to a detector and a L2 issued to a detector to be about 450 us in
the first 10000 evts and 350 us in the rest. Here and there, I see an
excursion to 10 times as much...
-- Tonko
-- Tonko (Ante) Ljubicic Email: tonko_at_bnl.gov Bldg. 510A Tel: +1-631-344-7346 Brookhaven National Lab., P.O. Box 5000 Fax: +1-631-344-4206 Upton, NY, 11973
This archive was generated by hypermail 2.1.4 : Thu Jul 24 2003 - 00:39:37 EDT