home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!gatech!destroyer!ncar!noao!arizona!arizona.edu!telcom.arizona.edu!leonard
- From: leonard@telcom.arizona.edu (Aaron Leonard)
- Newsgroups: comp.os.vms
- Subject: Re: Vax to Ultrix delay over decnet
- Message-ID: <1992Jul22.135053.3523@arizona.edu>
- Date: 22 Jul 92 20:50:47 GMT
- References: <1992Jul22.170955.22876@macc.wisc.edu>
- Reply-To: Leonard@Arizona.EDU
- Distribution: usa,local
- Organization: University of Arizona Telecommunications
- Lines: 54
- Nntp-Posting-Host: bugs.telcom.arizona.edu
-
- In article <1992Jul22.170955.22876@macc.wisc.edu>,
- stevens@vms.macc.wisc.edu (PAul STevens - MACC - 2-9618) writes:
-
- >> When we dlogin from ultrix to vms, all works well. Going the
- >>other way, vms to ultrix (via SET HOST), we experience really long
- >>delays [....]
- >
- >I have found that this happens when an ethernet packet is lost. The
- >decnet timeout value is 6 seconds. Can't be changed as far as I can
- >tell.
-
- See the NCP commands SET EXEC DELAY FACTOR and SET EXEC RETRANS
- FACTOR. The DELAY FACTOR is the minimum time (in sec/16) that
- DECnet NSP will wait before retransmitting a packet. By default, it's
- set to 80, and so NSP will wait 5 seconds before retransmitting.
-
- >Telnet's timeout is generally a few tens of milliseconds.
-
- Not true. If it were, then every high- or variable-delay network
- would quickly be destroyed by TELNET. TCP uses a sophisticated
- adaptive retransmission mechanism, but RFC-1122 recommends
- beginning with a retransmission timeout of 3 seconds.
-
- >Decnet must have been manufactured in a clean-room with three nodes.
- >It works rather poorly on busy, noisy networks. I have fought these
- >problems for years and all I can ever get from DEC is advice to make
- >the ethernet work better. IP protocols are much more robust. I have
- >seen cases where over 80 percent of packets sere being lost (no---that
- >is not normal, even here!) and telnet users had no complaints. VAX
- >clusters simply crashed.
-
- Your TELNET users must have the patience of Job, then. Performance on
- all network protocols that I am familiar with will go right down the
- toilet if you have as much as 2% packet loss, let alone 80%.
-
- It is true that LAVc nodes will bugcheck if they can't find the cluster for
- RECNXINTERVAL seconds, which is set by default to 20 (I recommend
- changing this to 120 on a large multipurpose network.)
-
- DECnet NSP is a relatively robust transport protocol, although not quite
- as massively tweaked as TCP. However, it performs much better in
- wide-area / lossy networks than, say, NFS, IPX, AppleTalk and many others.
-
- A properly configured Ethernet will lose virtually no packets (<0.1%)
- until it's grotesquely overloaded (>>90%). Greater packet loss than
- that is sign of electrical misconfiguration or hardware malfunction.
- Find someone responsible for your network, and encourage him to
- fix it.
-
- Aaron
-
- Aaron Leonard (AL104), <Leonard@Arizona.EDU>
- University of Arizona Network Operations, Tucson AZ 85721
- |> If it ain't broke, break it.
-