Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP Path: utzoo!mnetor!uunet!seismo!gatech!bloom-beacon!think!ames!sdcsvax!ucbvax!EMBL.BITNET!OMOND From: OMOND@EMBL.BITNET (Roy Omond) Newsgroups: comp.os.vms Subject: Info wanted on SIlink Message-ID: <8708172047.AA09511@ucbvax.Berkeley.EDU> Date: Mon, 17-Aug-87 12:41:04 EDT Article-I.D.: ucbvax.8708172047.AA09511 Posted: Mon Aug 17 12:41:04 1987 Date-Received: Tue, 18-Aug-87 05:58:04 EDT Sender: daemon@ucbvax.BERKELEY.EDU Distribution: world Organization: The ARPA Internet Lines: 42 Fellow Netters, Would anyone out there have any experience of SIlink from System Industries ? It's claimed that SIlink can enable one to build VAXclusters without the need for HSC and CI; This would be a typical SIlink cluster : ---+------------+-----------------------+--------- Ethernet | | | uVax-II uVax-II .... any Vax (up to 8 nodes) \ \ / \ \ / DMA connections \ \ / +-----------------------------------------+ ! SI 9900 controller ! +-----------------------------------------+ / / \ / / \ / / \ SI disk SI disk .... SI disk (up to 8 big disks) Basically, how SI claim this all works is that all the distributed lock manager traffic takes place over Ethernet, but all data traffic is done DMA through the 9900 controller (cf. DEC LAVC where the Ethernet becomes a bit of a bottleneck). It is claimed that the performance of this can approach that of a "real" cluster, certainly that it outperforms a LAVC. (Note that this is not SIMACS, which seems to have died a death, not before time :-) If I receive enough replies, I'll summarize to the net ... Roy Omond System Manager etc. European Molecular Biology Laboratory Heidelberg, West Germany. Disclaimer: I have nothing whatsoever to do with SI