NetNews Usenet Archive 1992 #19/NN_1992_19.iso/spool/comp/protocol/nfs/2235

Internet Message Format | 1992-09-04 | 3.8 KB

Xref: sparky comp.protocols.nfs:2235 comp.protocols.tcp-ip.ibmpc:5060 comp.dcom.lans.ethernet:1826
Path: sparky!uunet!ogicse!das-news.harvard.edu!spdcc!dirtydog.ima.isc.com!newsserver.pixel.kodak.com!tgl
From: [e-mail] (Tom Lathrop)
Newsgroups: comp.protocols.nfs,comp.protocols.tcp-ip.ibmpc,comp.dcom.lans.ethernet
Subject: Sun <=> PC Transfer Rate (Summary)
Message-ID: <[e-mail]>
Date: 4 Sep 92 17:39:31 GMT
Article-I.D.: pixel.1992Sep4.173931.24033
Sender: [e-mail]
Organization: Eastman Kodak
Lines: 66
Originator: tgl@jupiter

I was asked to summarize the responses to my query about optimizing
Sun to PC file transfer.

First, many thanks to all who responded.

I said:

>I would like to find a fast way to transfer large files from a Sun
>to a 486 PC. Currently we're using Sun PC-NFS over Ethernet, but we're
>not locked into that. Something like FTP, or writing our own custom
>program using sockets, would be acceptable. The files are tens or
>hundreds of megabytes. Transfers are mostly one-at-a time, on a dedicated
>network (essentially no other network traffic). Any suggestions would be
>appreciated.

The consensus seems to be that FTP is faster than NFS for file transfer.
FTP is included with the Sun PC-NFS package, although several respondants
preferred other implementations, including NCSA (available by
anonymous FTP from zaphod.ncsa.uiuc.edu) and Erick Engelke's implementation
for the University of Waterloo, available by FTP from 129.97.50.50 in
pub/wattcp/ftp.zip. Probably a good FTP implementation is fast enough
that a custom program wouldn't be much faster.

>    - How much difference does the PC Ethernet card make? Which board(s)
>      give the best performance?

There was general agreement that cards make a great deal of difference.
16 bit cards are preferable to 8 bit cards, and a large buffer is quite
helpful. Boards (and supporting software) which support shared memory
may be desirable, but a bus-master card should be even faster (EISA and
MCA busses only).

>    - Is there another vendor of NFS for the PC which has a product with
>      a better transfer rate?

None of my respondants had much to say about this, but the latest
(Sept. '92) issue of _SunWorld_ magazine has a test of 5 packages,
including Sun's PC-NFS, with performance numbers. Their best time
for Sun to PC transfer was using Wollongong's Pathway Access 2.0
and FTP. (Pathway's FTP was the slowest for PC-to-Sun transfer, though.)
It took 7 seconds to send a 1 megabyte file from a SPARCstation 2 to
a Dell 386/33 with a 3COM 3C503 board. This works out to 143k bytes/sec,
not very good. But 3COM and others make much better boards than the
3C503. Locus Computing's PC-Interface 4.10 scored the best on NFS
performance, but they don't support FTP.

>    - Currently we're using a SparcStation 2. Would a more powerful
>      machine, or one with better I/O, give better performance? Or
>      is there enough of a bottleneck on the PC end that it wouldn't
>      make much difference?

There was a strong consensus that the PC is the bottleneck. Two
SPARCstation 2's are apparently capable of data rates of 800 to 1000
kilobytes per second memory-to-memory transfer on a dedicated Ethernet.
In addition to the Ethernet card, the PC backplane bus was seen as a
potential bottleneck. EISA and Microchannel may well make a difference
for this type of application. (You can transfer 32 bits at a time, rather
than 8 or 16. Also, some EISA and MCA cards can become bus masters, allowing
incoming data to be transferred directly to main memory without CPU
intervention.) Also, the PC disk drive may be a bottleneck unless it has
a fast transfer rate. The CPU is probably not critical, though. On the
software side, a couple of people said that if a packet driver is used,
which one it is makes a big difference.

Tom Lathrop
[e-mail]