Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP Path: utzoo!watmath!clyde!burl!ulysses!allegra!mit-eddie!genrad!panda!talcott!harvard!caip!seismo!umcp-cs!cvl!umd5!louie From: louie@umd5.UUCP (Louis Mamakos) Newsgroups: net.lan Subject: Re: Ethernet connections for mainframes Message-ID: <930@umd5.UUCP> Date: Wed, 23-Apr-86 22:04:54 EDT Article-I.D.: umd5.930 Posted: Wed Apr 23 22:04:54 1986 Date-Received: Sun, 27-Apr-86 07:16:25 EDT References: <1104@uwmcsd1.UUCP> <169@brl-sem.ARPA> <926@umd5.UUCP> <774@osiris.UUCP> Reply-To: louie@umd5.UUCP (Louis Mamakos) Distribution: net Organization: U of Md, CSC, College Park, Md Lines: 58 In article <774@osiris.UUCP> eric@osiris.UUCP (Eric Bergan) writes: >> >We would like to hear from other sites that have such a configuration [mainframe >> >(of any sort) to Ethernet], preferable by means of an IO channel connection, >> >since our application requires the bandwidth of an IO channel. >> >> Yes, as far as I know, we were the first with TCP/IP on the Sperry. We're >> currently running a 40KB sync line to a fuzzball which has an ethernet on it. >> It seem that we've got some users, and they have all this data you see.. and >> they want to move it REAL fast. So. We are in the process of evaluating >> the Sparticus (now Fibronics) and Auscom IBM-type-byte-channel ethernet >> boards, and hope to have one here soon. Well see how fast our purchasing >> people can deal with this one. > > We have an Auscom currently connected to our IBM systems, running >XNS. I believe that our Auscom is a block channel, not a byte channel, >interface, though. The XNS resides in the Auscom itself, and the >virtual circuits look like a series of 327x terminals. This avoids the >protocol load on the host. We have implemented Sun's RPC mechanism in >the IBM, and our IBM is routinely making RPC calls from COBOL to our >Pyramid over all of this. (The Pyramid does database lookups, and >returns the information to the IBM.) We presented a paper on all of >this at the last Uniforum, for those that are interested in more >details. > > We have also looked at the Sparticus equipment. At the time we >were starting, Sparticus only supported VM, not MVS. I understand that >this is currently changing. There is, however, still the issue that >Sparticus runs TCP/IP in the host, and so you still have to pay for the >mainframe cpu cycles to handle the protocol. We would be very >interested in any Sparticus customers that could tell us how much of >their cpu gets chewed up doing the TCP/IP protocols. I remember that >the old UCLA code was estimated at chewing up about a third of a >reasonable sized IBM mainframe, which is far more than we could put up >with. > >-- > > eric > ...!seismo!umcp-cs!aplcen!osiris!eric I guess my terms were a bit incorrect. When I said 'byte' channel I meant an IBM style byte or block mux channel on the Sperry rather then the Sperry 36 bit wide "word" channel. We've been running TCP/IP on the host for quite a while, and found that it doesn't present a high enough load to worry about off-loading the processing to a front end. For our application, we'd prefer not to have an overly clever front end "helping" us, but rather a simple interface to an ethernet that won't bend the existing code too far out of shape. We'll see what happens; usage patterns are likely to change when the higher bandwidth network connection arrives. Who knows what these crazy users will do next! -- Louis A. Mamakos WA3YMH University of Maryland, Computer Science Center Internet: louie@trantor.arpa UUCP: {seismo!umcp-cs, ihnp4!rlgvax}!cvl!umd5!louie