[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Date: Tue, 15 Jun 1993 13:31 PDT
From: Barry Margolin <barmar@Think.COM>
Date: Tue, 15 Jun 1993 14:35 EDT
Restoring world from cdrom mounted on a sun server works like a charm
for 3670 and 3640 using copy file (symbolics recommended procedure).
Seems to run ~100 times slower on 3650 -- in fact, not enough time in a
day to let it complete. Eventually pops up a restart menu: longer timeout,
retry connection, etc.
Symbolics tried swapping io board, paddle, and processor. Tried swapping
ethernet connection with one of the machines that works okay. Nothing
helps. Any known problems with 3650 in regards to this? The only thing
I can think of is that NSF is sending a lot of long packets and the 3650
has a slow(?) fep.
Yes. The problem is that the default read size is 8K, and this is
fragmented into 6 IP packets sent in rapid succession. Unfortunately, a
G-machine can only read a couple of them, so the reassembly fails and
the Lispm eventually has to request a retransmission. Of course, this
also fails, and so on.
Ths solution is to set NFS:*LOCAL-SUBNET-TRANSFER-SIZE* and
NFS:*LOCAL-NETWORK-TRANSFER-SIZE* to something the Lispm can handle. I
use 1024 and it works OK.
That makes sense, but as a practical matter, I'd use :Copy World after the first one.
Both the disk access and network should be faster.