[Simh] Cluster communications errors

[Mark Pizzolato]

There are no deliberate buffering delays in the Ethernet layer.  There merely is one thread which receives (and filters) packets and queues the potentially interesting ones.  The available queue gets drained as fast as the simulated system happens to read the available data.  This might affect some worst case situations, but overruns due to speed mismatches and limited capacity of the old physical hardware are much more likely to blame.  Like I said, I’ve got multiple simulated LAVC nodes that can all talk just fine without the errors Hunter is seeing which if Bufferbloat was a factor might be worse there…


-          Mark

From: Simh [mailto:simh-bounces at trailing-edge.com] On Behalf Of Warren Young
Sent: Wednesday, July 18, 2018 2:33 PM
To: simh at trailing-edge.com
Subject: Re: [Simh] Cluster communications errors

On Wed, Jul 18, 2018 at 12:21 PM Mark Pizzolato <Mark at infocomm.com<mailto:Mark at infocomm.com>> wrote:

The simh Ethernet layer has dramatically more internal packet buffering (maybe 50 X) than anything real DEC hardware ever had.  This might account for the relatively smooth behavior I’m seeing.

More buffering can also mean more delay in the feedback loop that controls the underlying protocols, leading to *worse* performance as buffer space goes up.

This is called Buffer Bloat in the TCP sphere:

    https://www..net/<https://www.bufferbloat.net/>

Perhaps the low-level protocols involved in VAX clustering have the same issue? They may be expecting to get some kind of feedback response, which is getting delayed through the buffering, which causes the real VAXen to kick the fake one out, thinking it's gone MIA.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.trailing-edge.com/pipermail/simh/attachments/20180718/9d096a3e/attachment-0001.html>