Herb Peyerl <Herb.Peyerl@sidney.novatel.ca> writes:

> The symptoms I am seeing (with if_ed) are that small transactions are 
> fine (mount, ls), but actually copying files (from a SunOS client to 
> a NetBSD/i386 server) will cause a hang.

Well, a preliminary report on what I just did.  The box I'm doing this on
is a 386DX-25 ISA machine, 8MB of RAM, with a 3C509 and a WD8013EPC (i.e.
an SMC Elite/16).  I've been using both cards with a SunOS 4.1.x NFS
server reliably for the last several days (3C509) and weeks (WD8013).  No
problems there.  I have had NFS_SERVER compiled into my kernel for a while
now, just in case, and decided to start up mountd/nfsd and mount a local
disk to the Sun IPX. 

With if_ep, I mounted the disk on the Sun and got as far as a single cp 
command which failed due to permissions being set wrong.  About a minute 
later, I got messages on the NetBSD box saying that the Sun had gone 
away, and then bootpd complained about no buffer space.  Pings to the 
NetBSD machine didn't get through.

With if_ed, I mounted the disk on the Sun and did two concurrent copies 
in different windows.  The first was "cp emacs* /mnt", the second "cp 
gcc*.gz /mnt" (I mounted the NetBSD disk on the Sun's /mnt).  To give you 
an idea of the file sizes,

-rw-r--r--  1 scott  staff  6045916 Sep  4  1993 emacs-19.19.tar.gz
-rw-r--r--  1 scott  staff  1873066 Feb 14  1993 gcc-1.42.tar.gz
-rw-r--r--  1 scott  staff    14726 Jan 31 00:51 gcc-2.5.7-2.5.8.diff.gz
-rw-r--r--  1 scott  staff  5955006 Jan 31 01:22 gcc-2.5.7.tar.gz
-rw-r--r--  1 scott  wheel    52653 Jan 19  1993 gcc1.39-1.40.diff.gz
-rw-r--r--  1 scott  wheel    69855 Jan 19  1993 gcc1.40-1.41.diff.gz
-rw-r--r--  1 scott  wheel     7185 Jan 19  1993 gcc1.41-1.42.diff.gz

This took a little while, as you might imagine.  My poor little machine got 
ground into the dirt.  It made it through, though, and when I eventually 
got a prompt back in each window I cd'ed to /mnt, dir a "mkdir tmp" and 
then "cp *.gz tmp" to pound on NFS in both directions.  The long and 
short of it is that the if_ed driver handled it just fine, if you don't 
mind waiting a bit longer than with the if_ep driver.

> Rumor has it this only affects slow machines (I have a 386-40).

What I've seen would appear to support this statement.

I don't have enough disk on my NetBSD boxes, so this isn't really a 
problem for me.  Just hoping to add some useful noise to the discussion.

--scott

------------------------------------------------------------------------------