wyle@ethz.UUCP (Mitchell Wyle) (11/01/87)
Is anyone else trying to boot a mac-II diskless like a Sun-3? -- Mitchell F. Wyle | csnet or arpa: wyle%ifi.ethz.ch@relay.cs.net Instituet fuer Informatik | uucp: wyle@ethz.uucp ETH Zentrum / SOT | Telephone: 011 41 1 256 5237 8092 Zuerich, Switzerland
verber@tut.UUCP (11/03/87)
The idea of a diskless MacII running A/UX has been *strongly* suggested to Apple. They held a meeting a few months ago with the universities who had recieved seed units to discuss possible improvements for A/UX. Everyone present stress that they would really like A/UX to boot via the network. This idea goes very much against the philosphy of computing that is present at Apple. They prefer the idea of stand alone machines that can share resources, i.e. being dependant on another machine for booting and all disk use doesn't sit well with them. On the other hand, they seems to hear the pleas of the the university people (who will be some of the biggest customers). I expect that we will see diskless A/UX in the next year or two. Apple hasn't committed to this, as far as I know, but they seem to understand why this will be important to succeed in the higher education market. Cheers, ----------------------------------------------------------------------- Computer Science Department Mark A. Verber The Ohio State University verber@ohio-state.arpa +1 (614) 292-7344 cbosgd!osu-cis!verber
phil@apple.UUCP (Phil Ronzone) (11/05/87)
In article <933@tut.cis.ohio-state.edu> verber@tut.cis.ohio-state.edu (Mark A. Verber) writes: > >The idea of a diskless MacII running A/UX has been *strongly* >suggested to Apple. They held a meeting a few months ago with the >universities who had recieved seed units to discuss possible >improvements for A/UX. Everyone present stress that they would really >like A/UX to boot via the network. This idea goes very much against >the philosphy of computing that is present at Apple. They prefer the >idea of stand alone machines that can share resources, i.e. being >dependant on another machine for booting and all disk use doesn't sit >well with them. On the other hand, they seems to hear the pleas of >the the university people (who will be some of the biggest customers). >I expect that we will see diskless A/UX in the next year or two. >Apple hasn't committed to this, as far as I know, but they seem to >understand why this will be important to succeed in the higher >education market. >Computer Science Department Mark A. Verber >The Ohio State University verber@ohio-state.arpa >+1 (614) 292-7344 cbosgd!osu-cis!verber No, no, Mark - we are NOT against diskless A/UX systems. It is very simple - we didn't have the time (and lacked access to appropriate source code) to implement the "page serving" protocols that SUN uses. We have NFS and YP - but not the disk page protocols. I assure your that Apple Marketing, on a regular basis, begs for full diskless A/UX capability. The rational demands of Higher Ed are transmitted to Apple software engineering with great fervor. If I just had another 12 or so very very experienced UNIX programmers ... :-) Philip K. Ronzone A/UX Technical Manager MAIL: Philip K. Ronzone Apple Computer Mail Stop 27AJ 10500 N. DeAnza Blvd. Cupertino, CA 95014 UUCP: ...!{sun,voder,nsc,mtxinu,dual,unisoft}!apple!phil
edmoy@opal.berkeley.edu.UUCP (11/05/87)
In article <6633@apple.UUCP> phil@apple.UUCP (Phil Ronzone) writes: >No, no, Mark - we are NOT against diskless A/UX systems. It is very >simple - we didn't have the time (and lacked access to appropriate >source code) to implement the "page serving" protocols that SUN uses. >We have NFS and YP - but not the disk page protocols. I assure your >that Apple Marketing, on a regular basis, begs for full diskless >A/UX capability. The rational demands of Higher Ed are transmitted to >Apple software engineering with great fervor. If I just >had another 12 or so very very experienced UNIX programmers ... :-) > > >Philip K. Ronzone >A/UX Technical Manager Speaking as someone who has watched (from afar, thank god!) almost a thousand diskless Suns go out to the UC Berkeley campus and having heard of the many disasters that the so-called ND (network disk) protocol creates, I would advise against using ND. (I think even certain people in Sun have No ND buttons that they wear.) The main problem with ND is that on the server, the "disk" is just a huge file. If the "file" gets trashed, preventing the client from booting, you can't poke around on the server side to figure out what's wrong, since it's just one huge data file and not a hundred odd Unix files in a Unix file system. The only way to fix the ND partition is to copy a good one on top of it. Particularily in an environment where users are not all computer science majors, it too often occurs that users think that if the system doesn't respond for 10 seconds, it's time for L1-A (Reboot key sequence). This often trashes the ND partition, and the client is dead until someone can fix it. The diskless workstation is an interesting concept, but not without its problems. Disk-intensive applications will run like a snail, and large, memory intensive applications may run slowly also, due to swapping across the network. I think the best compromise is the minimum-disk workstation. Have a small disk with the essential stuff to boot-up and enough for a good size swap area. Then mount a remote filesystem where most of the work is done. You might even have a medium-disk workstation, which has some extra room for those times when the server is down, or for running disk-intensive applications. Edward Moy Academic Computing Services University of California Berkeley, CA 94720 edmoy@opal.Berkeley.EDU ucbvax!opal!edmoy P.S. Boy am I glad I have a "diskful" MicroVax II.
chuq@plaid.Sun.COM (Chuq Von Rospach) (11/06/87)
>>No, no, Mark - we are NOT against diskless A/UX systems. It is very >>simple - we didn't have the time (and lacked access to appropriate >>source code) to implement the "page serving" protocols that SUN uses. >>We have NFS and YP - but not the disk page protocols. I assure your >>that Apple Marketing, on a regular basis, begs for full diskless >>A/UX capability. >Speaking as someone who has watched (from afar, thank god!) almost a thousand >diskless Suns go out to the UC Berkeley campus and having heard of the many >disasters that the so-called ND (network disk) protocol creates, I would >advise against using ND. (I think even certain people in Sun have No ND >buttons that they wear.) I don't want to get into the good points and bad points of ND (I own three "No ND" buttons, personally...) but I do want to point out that Apple couldn't port ND if they wanted to, because ND isn't a technology that Sun make available to outside parties (unlike NFS, YP, RPC, XDR, and NFS). Of course, with SunOS 4.0, ND will be going away, and NFS enhanced so that it will take over all the functionality of ND, clean up some of the administrative glitches that ND users learn to love, and will remove any excuse for Apple to not support diskless machines (hee-hee). chuq --- Chuq "Fixed in 4.0" Von Rospach chuq@sun.COM Delphi: CHUQ
verber@tut.cis.ohio-state.edu (Mark A. Verber) (11/06/87)
Phil & Everyone, I am sorry that my posting gave the impression that *all* people at Apple were against diskless machines. There are some notable exceptions in the A/UX group, and elsewhere within Apple. But historically people within Apple have not been very responcive to the idea of diskless machines and centralize admin. I tried to convey that this seems to be changing, without saying that Apple was definitively going to release a diskless A/UX product, since that is not for me to say. Cheers, ----------------------------------------------------------------------- Computer Science Department Mark A. Verber The Ohio State University verber@ohio-state.arpa +1 (614) 292-7344 cbosgd!osu-cis!verber
phil@apple.UUCP (Phil Ronzone) (11/06/87)
In article <33128@sun.uucp> chuq@sun.UUCP (Chuq Von Rospach) writes: >I don't want to get into the good points and bad points of ND (I own three >"No ND" buttons, personally...) but I do want to point out that Apple >couldn't port ND if they wanted to, because ND isn't a technology that Sun >make available to outside parties (unlike NFS, YP, RPC, XDR, and NFS). > >Of course, with SunOS 4.0, ND will be going away, and NFS enhanced so >that it will take over all the functionality of ND, clean up some of >the administrative glitches that ND users learn to love, and will >remove any excuse for Apple to not support diskless machines (hee-hee). > Yeah - that's the way I should have phrased it in the first place. To duplicate ND would not have been worth it given what is coming down the pike. However, changing context a bit: If SUN 3/260 can really only handle 10-12 diskless before sluggishness overwhelms, then isn't the true cost of those 10-12 diskless workstations = workstation_cost + 1/10 or 1/12 of 3/260 cost? And what if a hard disk is much less than 1/10 or 1/12 of a 3/260, and said hard disk allows 20-24 workstations per server? I am very interested in the performance and cost tradeoffs in this area and would like input - it will affect Apple A/UX development priorities. ---- Philip K. Ronzone A/UX Technical Manager MAIL: Philip K. Ronzone Apple Computer Mail Stop 27AJ 10500 N. DeAnza Blvd. Cupertino, CA 95014 PHONE: (408) 973-2509 UUCP: ...!{sun,voder,nsc,mtxinu,dual,unisoft}!apple!phil
len@asterix (Len Lattanzi) (11/07/87)
In article <5790@jade.BERKELEY.EDU> edmoy@opal.berkeley.edu () writes: >The main problem with ND is that on the server, the "disk" is just a huge >file. If the "file" gets trashed, preventing the client from booting, you >can't poke around on the server side to figure out what's wrong, since it's >just one huge data file and not a hundred odd Unix files in a Unix file system. >The only way to fix the ND partition is to copy a good one on top of it. >Particularily in an environment where users are not all computer science >majors, it too often occurs that users think that if the system doesn't >respond for 10 seconds, it's time for L1-A (Reboot key sequence). This often >trashes the ND partition, and the client is dead until someone can fix it. > WRONG! I know this doesn't belong in comp.sys.mac but then again neither does mis-information. ND partitions can be treated like file-systems. See ND(4p). In brief, you can use /dev/ndl* (the server side of an ND partition) to run mkfs/fsck/mount....though the server should only do this when the client is halted. -Len ARPA:len%spar@decwrl.dec.com UUCP:decwrl!spar!len BELL:(408) 732-2343 USPS:MS 32-930 1601 Technology Drive, San Jose, CA 95110-1397 Schlumberger Technologies ATE will disavow any knowledge of my opinions.
edmoy@opal.berkeley.edu (11/07/87)
In article <6655@apple.UUCP> phil@apple.UUCP (Phil Ronzone) writes: >If SUN 3/260 can really only handle 10-12 diskless before sluggishness >overwhelms, then isn't the true cost of those 10-12 diskless workstations >= workstation_cost + 1/10 or 1/12 of 3/260 cost? > >And what if a hard disk is much less than 1/10 or 1/12 of a 3/260, and >said hard disk allows 20-24 workstations per server? > >I am very interested in the performance and cost tradeoffs in this area >and would like input - it will affect Apple A/UX development priorities. > ---- Of course it depends on what the users on the clinets are doing. If they login just to read their mail and do an occassional n/troff, then you could probably get by with 15-20 workstations on a server. On the other hand, we have a computer science class that runs some heavy-duty software and they were forced to run without SunWindows (because it is a real resource hog). They either ran the X window system or often no window manager at all (yuk!), so that the swapping and paging load didn't kill the clients or the server. Sun was supposed to be running some test of workstations with local disks for swapping and paging, running under heavy load. Has anyone heard the results? In my opinion, this is the way to do it (the minimum-disk workstation). Edward Moy Academic Computing Services University of California Berkeley, CA 94720 edmoy@opal.Berkeley.EDU ucbvax!opal!edmoy
cswarren@enzyme.berkeley.edu.BERKELEY.EDU (Warren Gish) (11/08/87)
In article <6655@apple.UUCP> phil@apple.UUCP (Phil Ronzone) writes: >If SUN 3/260 can really only handle 10-12 diskless before sluggishness >overwhelms, then isn't the true cost of those 10-12 diskless workstations >= workstation_cost + 1/10 or 1/12 of 3/260 cost? > >And what if a hard disk is much less than 1/10 or 1/12 of a 3/260, and >said hard disk allows 20-24 workstations per server? > >I am very interested in the performance and cost tradeoffs in this area >and would like input - it will affect Apple A/UX development priorities. You may be able to come up with an initially less expensive system that uses diskful clients (i.e., workstations with their own disk drives for swap and OS), but my impression is that the longterm cost will actually be higher, as in: (1) increased system administration costs -- OS upgrades cost more to install, and backup procedures may be more complex. (2) more parts to break, more equipment to insure, maybe more desk space given up. (3) a server with large disk drives is still a likely necessity. Often the bulk of the load on a server and the network these days is due to swapping on the clients. In the not-too-distant future, RAM should be cheap enough that swapping will be a minor issue. RAM caches will be larger, too. A Sun-3/50 has only 4 MB RAM, whereas 8+ MB clients may be common place in less than a year. I would discount somewhat the current experiences with 4 MB clients. My vote is for diskless. I'll also volunteer to beta-test a 128 MB Mac!
phil@apple.UUCP (Phil Ronzone) (11/09/87)
In article <5824@jade.BERKELEY.EDU> cswarren@enzyme.berkeley.edu.BERKELEY.EDU (Warren Gish) writes: >You may be able to come up with an initially less expensive system that uses >diskful clients (i.e., workstations with their own disk drives for swap >and OS), but my impression is that the longterm cost will actually be higher, >as in: > > (1) increased system administration costs -- OS upgrades > cost more to install, and backup procedures may be more complex. > > (2) more parts to break, more equipment to insure, maybe more > desk space given up. > > (3) a server with large disk drives is still a likely necessity. > >Often the bulk of the load on a server and the network these days is due >to swapping on the clients. In the not-too-distant future, RAM should >be cheap enough that swapping will be a minor issue. RAM caches will >be larger, too. A Sun-3/50 has only 4 MB RAM, whereas 8+ MB clients >may be common place in less than a year. I would discount somewhat >the current experiences with 4 MB clients. > >My vote is for diskless. I'll also volunteer to beta-test a 128 MB Mac! (1) Are the sys admin costs higher? Higher than the decreased overall per system cost? I envisioned a diskful workstation with a clone disk -- never essentially changing, or allowed to be changed by the user. It is just there to get executable loads and swapping off the net AND off the server. Do your economics cost out both the net traffic and server usage decrease? In article <5821@jade.BERKELEY.EDU> edmoy@opal.berkeley.edu () writes: >Sun was supposed to be running some test of workstations with local disks >for swapping and paging, running under heavy load. Has anyone heard the >results? In my opinion, this is the way to do it (the minimum-disk >workstation). Yes - I was informed of the results by the person at Berkeley who paid (that's right - paid SUN to run the tests) but was not given permission to copy or quote from them. The surprising thing to me at the time was the economics - in short, if your hard disks start costing the end-user under $5,000, then diskful starts making more and more sense, and the decrease in useless net traffic such as paging (O.K. - my biases ARE showing :-) ) is the cherry on top. Now since I am not in a position to either copy the report, or buy 3-4 servers and 60 or so workstations just to run a few scientific tests ( :-) ), I am wondering if any out there has. > (2) more parts to break, more equipment to insure, maybe more > desk space given up. Well, yes (not space, in a Mac II, the extra parts are a disk, a data cable, and a power cable inside the box). The other hand is that when your server breaks, your systems are still autonomous, or could be sent to another server without grossness (I've seen 10 additional diskless systems switched to another server when the first server went down. 33 diskless systems on 3/280? It didn't work too well). > (3) a server with large disk drives is still a likely necessity. I agree - SUN large servers are very nice boxes. And I'd like to make A/UX such that I can connect as many Mac II's to SUN servers as possible - but I am trying to accumulate facts (hard ones) on software priorities. Example, add even more sophistication to the A/UX SCSI driver to enhance even more its support of generic SCSI drives, OR, pick up ND-like protocols (or wait for whatever is in the next NFS release), or what? As I said, since I don't have the facilities to really test out the rule of thumb "facts" I've encountered, I'd very much like to find someone who has. And I also welcome strong feedback on what Apple should be doing with UNIX A.K.A. A/UX. Philip K. Ronzone A/UX Technical Manager MAIL: Philip K. Ronzone Apple Computer Mail Stop 27AJ 10500 N. DeAnza Blvd. Cupertino, CA 95014 PHONE: (408) 973-2509 UUCP: ...!{sun,voder,nsc,mtxinu,dual,unisoft}!apple!phil
rcopm@koel.rmit.oz (Paul Menon) (11/12/87)
in article <6655@apple.UUCP>, phil@apple.UUCP (Phil Ronzone) says: > If SUN 3/260 can really only handle 10-12 diskless before sluggishness > overwhelms, then isn't the true cost of those 10-12 diskless workstations > = workstation_cost + 1/10 or 1/12 of 3/260 cost? Depends where the (unbearable) sluggishness occurs - At the Workstations only? - At both the W/S's and for all Sun 3/260 users as well. My initial hunch would be "Sluggishness at the Workstations", as they would show greater performance degradation than would an equivalent no of local (Server) users. If the "server" can still be useable to local clients with 10-12 w/s's, then one cannot use the above formula. I would also be interested in where the 10-12 figure came from, as the faculty here is looking into ND's. > > And what if a hard disk is much less than 1/10 or 1/12 of a 3/260, and > said hard disk allows 20-24 workstations per server? I don't understand the above statement. How will a hard disk allow 20-24 w/s per server, and surely in this case those w/s are N/Ds? Unless we are talking about "local" w/s (terminals). A couple of points worth mentioning... * The most unreliable part of computers here (hic city) are hard disks. Well, considering hardware only, software and people are exempt. Disk (hardware) maintenance is expensive. So is disk s/w maintenance as in backup/recovery. The more disks - the more trouble we have. It is easier to look after fewer disks. Software upgrades are easier and quicker to implement. * The above is from the end user point of view of course. Why would Sun people wear NO ND buttons? more hunches... * Sun Disks are expensive. Getting advice from them about using third party disks is like asking Apple the same question - Do It at Yer Own Risk! * With Sun OS 3.3, a 20 line grahics application (SunView) takes up 1/2Mbyte of disk space. Don't ask how much memory. Sure they have been talking about shared libraries, etc - but I haven't seen them yet. At least Apple have their grahics routines in ROM. "Any Opinions Found Above Do Not Reflect - They Have Been Sandblasted" Paul Menon. Dept of Communication & Electronic Engineering, Royal Melbourne Institute of Technology, 124 Latrobe St, Melbourne, 3000, Australia ACSnet: rcopm@koel UUCP: ...!seismo!munnari!koel.rmit.oz!rcopm CSNET: rcopm@koel.rmit.oz ARPA: rcopm%koel.rmit.oz@seismo BITNET: rcopm%koel.rmit.oz@CSNET-RELAY PHONE: +61 3 660 2619.