MWC's Coherent - A Lemon...

[Hwa Jin Bae]

In article <2803 at crash.cts.com> jca at pnet01.cts.com (John C. Archambeau) writes:
   Try mixing Sun with an Appletalk to ethernet gateway and throw in a few PC's
   Also, I don't have any idea what you 'hacked' into your realtime OS.  If
   reasonable.  Or even better yet, try throwing in a network of Macs with
   ethernet boards.

Wrong assumptions. The network I have here does include a PC running
PC NFS and a Gatorbox Appletalk gateway and various other vendors'
machines, in addtion to the ~30 Sun Workstations and a few file
servers.  We also have multiple Macs with ethernet boards on the
network.  I'm not sure what you're trying to prove here; are you
saying that PC's and Mac's will affect the throughput limit of a
segment of ethernet?

   Also, I don't have any idea what you 'hacked' into your realtime OS.  If
   ethernet performance was a high priority, then you obviously did some fudging
   to bump up the performance.

Nope.  Wrong again.  The modifications were made to the original code
to make it work with our OS.  Many minor modifications were made including
rewriting checksum code, rearchitecting the mbuf and clustering mechanisms,
recoding a lot of the redundant socket level code, optimizing various
ethernet drivers to minimize copying (loaning out LANCE ring buffers
to eliminate an extra bcopy), etc.  The priority of the network task doesn't
need to be bumped up at all.  In fact, the priority of the network task is
quite low, consuming about less than 50 percent of the available CPU time
on a typical MC68020's.  The issue you were concerned about was the ethernet
throughput.  Whatever the point you're trying to make here (I'm not sure
what your point is), try to perform some real tests and write some real code
to verify your thesis, come up with some real data, analyze the data and
let us know.

hwajin


--
hwajin at daahoud.wrs.com