Thread View: comp.protocols.tcpip
2 messages
2 total messages
Started by john.sottile
Tue, 07 Jan 1992 07:27
Using NFS over dialup Lines (SLIP/PPP)
Author: john.sottile
Date: Tue, 07 Jan 1992 07:27
Date: Tue, 07 Jan 1992 07:27
77 lines
4417 bytes
4417 bytes
We have been experimenting with NFS over SLIP and PPP
using Telebit's NetBlazers with 2 t3000 modems. Here are some
observations and NFS configurations that we used to optimize NFS. It
may help some folks out there.
Equipment:
SUN Sparc 1, 1+, 2
SUN 4/390 File Server
4 t3000 Modems (2 on each end)
2 Telebit NetBlazers (2 port Serial Card on each)
Telebit's 1.20 NetBlazer Software
We Run SLIP over these Dialups.
We tuned NFS via the /etc/fstab options to use a retrans=300,
rsize=4096,wsize=4096.
Read Size and Write Size = 4K (we found that 2K per t3000 modem pair
works nicely)
Retransmission = 300 tenths of a second (30 Seconds). This is a bit
excessive, but our NFS File Server has 12 NFSD and rarely looses a
packet. We will wind up backing this down to about 20-25 seconds if
the 30 second timeout is too long.
Since SUN's NFS uses RPC (which uses udp), you can use
/usr/etc/nfsstat -c
to find out how many packets are retransmitted. Still further,
you can read the
MANAGING NFS and NIS from O'Reilly and associates; the Performance
Chapter to tune for the optimum performance for your dialup/slow speed
connection.
This book, along with almost every other O'Reilly Book, is fantastic
in helping to deal with day to day problems. This is the best book on
NFS and NIS that I have seen yet (no, I am not associated with
O'Reilly).
Tuning the MTU and TCP Window Size on the NetBlazer s also help to
keep fragments down. The 2K rsize per modem helps to keep the
NetBlazer's queue full. With 2 pairs of modems (2 phone lines), 2K
per modem pair helps to keep both modems pushing data. The MTU we use
is 1500 on the netBlazers (almost the same as Ethernet), the TCP
Window Size we use has varied from 4K to 8K. This doesn't help NFS,
but it does help TCP/IP packets. Using 4K for 2 modems causes the
NetBlazer to fragment the data and send the fragments over both
modems. Telebit suggests 512 byte packets for SLIP, but the TCP/IP
traffic is high and the modem isn't kept as busy thus some lost
bandwidth. The key here is the retrans. NFS from SUN is implemented
with RPC and winds up being a farily synchronous implementation.
Therefore, with a large rsize (the rsize for SunOS is 8K, but this is
too large for dialup) and a large retrans, you can keep the modem
pretty busy.
With the default configuration of NFS, over a dialup we found 80%
duplication in packets. This is mostly due to NFS being tuned on the
SUN for 10MB ethernet. With the above options, we found less than 5%
duplicate (mostly in the 2% range) which is acceptable. With the 2
port Serial card on the NetBlazers, asy00 + asy01 < 56Kbps. This
holds true since we see between 4.5K/s and 5.3K/s transferred with NFS
over dialup. Telebit's 1.40 netblazer OS will allow 56K per asy
allowing us even more performance. The 8 Port "DIGI" Serial Card (NM8
from Telebit) allows 56K per asy onthe 1.20 version.
Anyone attepting to do NFS over SLIP/PPP could benefit from similar
improvements. Most NFS implementations have these type of "tunables"
Even those doing NFS over 56K to T1 speed lines can benefit from
tuning NFS in a similar way.
Just some information for anyone interested. Anyone else have any
further possibilities, drop me a line.
=============================
John Sottile "Home is where the Software Grows"
AT&T Creative Software #include <std/disclaimer/who-me?.h>
jjs@addams.att.com
Re: Using NFS over dialup Lines (SLIP/PPP)
Author: Jim Rees
Date: Tue, 07 Jan 1992 17:04
Date: Tue, 07 Jan 1992 17:04
11 lines
1465 bytes
1465 bytes
In article <1992Jan7.072754.10321@cbfsb.att.com>, eajjs@cbnewsf.cb.att.com (john.sottile) writes: We have been experimenting with NFS over SLIP and PPP using Telebit's NetBlazers with 2 t3000 modems. Hard to believe anyone would try to run NFS over a slow link. It doesn't do any client-side caching, does it? How do you deal with that? We're running AFS over SLIP here and it seems to work pretty well (after tuning rx a bit). Our client caches are about 20 Mb and are persistent across reboots.
Thread Navigation
This is a paginated view of messages in the thread with full content displayed inline.
Messages are displayed in chronological order, with the original post highlighted in green.
Use pagination controls to navigate through all messages in large threads.
Back to All Threads