PPP/Net problems in recent kernels

Bill Earnest (wde@fast.net)
Fri, 22 Mar 1996 07:53:08 -0500 (EST)


Hello,
This is a summary of a series of subjective tests I have tried on
recent kernels to narrow down the problem(s) affecting the network. First
a system summary: 486DX2-66, 32MB RAM (everything ran in memory, the swap
partition use never got above 0), IDE & SCSI drives, latter off a FD-950
dumb controller, 28.8 modem usually connecting at 24K, interfaced to
serial 16550A at 38400. Using PPP-2.2.0e, which has given no problems.
Test case usually the startup of xrn from the news server at my ISP, which
involves the download of the ~380k index to about 15000 news groups.
Monitoring of performance via pppmond/xpppmon and sometimes via pppstats.

Up thru 1.3.72, performance was "normal", with the xrn startup
taking about 3.5 minutes, and xpppmon showing a steady packet rate limited
only by the modem link, with an occasional minor glitch caused by a lost
packet every 60-90 seconds. This loss has been chronic, and coincides with
disk activity (update) due to a known latency problem in the SCSI area.

For kernels .73 & .74, packet loss got significantly worse, with a
lost packet every 2-10 seconds. The resulting timeout-retransmit from my
ISPs server increased the startup time by over 50%, but the link was still
usable locally. This loss level made high delay (overseas) connections
practically impossible. The xpppmon trace never got up to the normal
thruput levels, but did show packets flowed most of the time. There was no
disk activity or other processes running that correlated with the loss
timing.

Kernels .75 thru .77 then became much worse, with packet losses
increasing to about 35% of all packets. Only the shortest mail items could
get thru, and these taking over 10* the normal time. An attempt to use xrn
gets nowhere, with delays between retries quickly extending out to
minutes. Only an occasional small spike appears on the xpppmon display,
thruput is essentially zero. As best I can estimate, the loss rate of
kernels .75 thru .77 is about the same.

All the above comes from compiling kernels .70 thru .77, using the
same config for each, and trying them alternately at a time that the ISPs
server was lightly loaded. If anyone can spot a connection or clue, please
do so...

Bill Earnest wde@fast.net
Fax (610) 432-7477