From:	CRDGW2::CRDGW2::MRGATE::"SMTP::CRVAX.SRI.COM::RELAY-INFO-VAX" 29-SEP-1989 21:55
To:	MRGATE::"ARISIA::EVERHART"
Subj:	Re: TCP/IP for VAXen

Received: From TGV.COM by CRVAX.SRI.COM with TCP; Fri, 29 SEP 89 16:04:06 PDT
Date:     Fri, 29 Sep 89 16:03:26 PDT
From: adelman@TGV.COM (Kenneth Adelman)
Reply-To: Adelman@TGV.COM (Kenneth Adelman)
Message-Id: <890929160326.20200061@TGV.COM>
Subject:  Re: TCP/IP for VAXen
To: ubc-cs!alberta!calgary!ctycal!ingoldsb@beaver.cs.washington.edu
Cc: info-vax@kl.sri.com

>>	 We are in the process of connecting our VAXen with the
>> rest of the Unix workstations.  From the net, I've heard
>> different opinions regarding Wollongong's and Multinet's
>> implementation of TCP.  Besides these two companies,
>
> We tried several and found we preferred Wollongong.  Make certain you
> test the product yourself.  Some TCP/IPs don't like to talk to other
> TCP/IPs.  Strange, but true.	Also, if you are interested in NFS be
> aware that most vendors offer only an NFS Server for VMS.  I believe
> Wollongong offers client/server.  My experience with VMS NFS is that
> it is *very* slow.  Do others share this opinion?  (I haven't tried
> Wollongong NFS).

    We (TGV) have an NFS Server as a product (part of MultiNet) and an
NFS Client under development.  I don't know of any other vendors which
have made any indication about having a client.  Catch us at Interop,
we'll have our alpha-client along to demonstrate.

    If your experience with VMS NFS is that it is very slow I suggest
you try some other vendors.  Ours, CMCs, and I believe DECs and
Excelan's LANService all offer acceptable performance (>100Kbyte/sec).
To give you some numbers, on a moderately loaded VAX-11/780 running
MultiNet over an ENP-40 ethernet card (the DEC cards become a
bottleneck if you do your NFS right) we see 310Kbyte/sec transfer
rates for read operations.  This is faster than a lot of SUNs (and we
do UDP checksumming ;-)).

    VMS NFS doesn't have to be slow.  One of the important design
issues in an NFS Server for VMS is whether the server is serial (i.e.,
only processes one request at a time) or whether it uses the
fully-asynchronous nature of the VMS filesystem to process requests in
parallel (from this point-of-view, VMS is a better operating system
than Unix for implementing an NFS Server).  Our first crack at an NFS
Server for VMS (never released to the public by us) was a port of the
SUN Unix reference port and ran serially.  We found that performance
was poor, and unacceptable when more than one client was accessing it
at the same time.  In our server as part of MultiNet we did everything
asynchronously, and normally handle up to 100 requests active at any
one time.  This, combined with creating a special interface between
the server and the network to minimize data copies (our server runs in
kernel mode and accesses the network directly) resulted in a server
with acceptional performance.

						    Kenneth Adelman
						    TGV, Inc.