Subject: com silo overflows go away with processor upgrade
To: None <current-users@NetBSD.ORG>
From: John C. Hayward <John.C.Hayward@wheaton.edu>
Date: 12/11/1996 08:09:30
I have since 1.0 been experienceing com silo overflows between two
NetBSD machines connected via ppp using USRobotics 28.8 modems in
compression mode. These silo overflows caused marked decrease in transfer
rates as corrupted packets would be lost and time to resync slowed the
process. I had reported there seemed to be a strong correlation between
the allocation of new disk space and the silo overflows in that if the
recieving file was /dev/null the frequency of overflows went down.
I had adjusted the com driver for the fifo trigger in the 16550
chip to trigger when the fifo dropped to 14 slots which helped in reducing
the com overflows.
Both machines were 486/33MZ machines.
Last night I upgraded the mother board on the machine which normally
recieves data to a P5/133MZ and upgraded memory from 8MB to 16MB.
After the upgrade I have not observed a silo overflow and transfer rate
of a binary like /netbsd was 4109 bytes per second. I am happy.
It appears that the silo overflow problems are related to either too
long sections of code in the operating running with interupts disabled
(at least from the com drivers persective) or processing of the com
interupt taking too long for a 486/33MHZ