info-vax@ucbvax.ARPA (11/13/84)
From: Jerry Wolf <WOLF@BBNG> Here, 2nd hand, is the answer I got when I asked a Deccie from their Storage Systems Group "why is DEC pushing the Unibus disks now? Why aren't the Massbus disks better?" When the VAX was designed, they needed a way to accomodate the instantaneous (intermittent) bursts from a disk (as high as 2.2 MB/s from an RP07) -- hence the Massbus, a big fast dumb pipe that's idle most of the time, between disk blocks. Also, they're so dumb that if multiple disks are ready, only one at a time can be active. Technology marches on, specifically enter cheap fast static RAMs and 2901 bitslice microprocessors, which made it feasible to build the UDA50 controller that can buffer the data off the disk at disk rates and send it to the CPU at Unibus rates, and handle 4 drives at once. The average throughput is held to be comparable to the Massbus (in most cases) -- the saving comes from not having to accomodate the high burst rate. Also, the UDA50 is supposed to be smart enough to reorder requests to multiple drives in order to minimize latency and maximize throughput, etc. This claim of better functionality through more advanced technology seems borne out by the fact that Unibus disks are a fair bit cheaper than Massbus, and actual thruput measurements (not just maximum transfer rates) are comparable in most cases, tho there are pathological applications where a Massbus disk can pump data faster.
info-vax@ucbvax.ARPA (11/13/84)
From: Bill Mitchell <whm%arizona.csnet@csnet-relay.arpa> Thanks for all the notes about the Unibus vs. Massbus disks; please keep them coming in if you have something to say. Let me pose a hypothetical question: If you were putting disks on a new 785 that was going to run 4.2bsd and you had to pick between a UDA50 and 2 RA81s or a SI 9900 and 2 9751s (Eagles) and there was no cost difference, which would you pick? Note that this machine would have a DEUNA and one UBA. Many thanks, Bill Mitchell
info-vax@ucbvax.ARPA (11/13/84)
From: lhasa!stew@harvard.ARPA The hypothetical question that you pose is impossible to answer. The Eagle and RA81/UDA50 configurations have quite different characteristics, and the answer is heavily dependant on the type of job you are running. The advantage of the UDA50 is that it does quite a lot of optimization: seek ordering, rotational latency, buffering, etc. In an environment which consisted of, say, several heavy page fault bound processes (for example, large lisp or other GC-based systems) this optimization will have a large effect. In other environments (say, database applications) which do larger bursts, the cpu-controller interface becomes the bottleneck and the Eagle/Massbus emulator configuration wins. The above is probably still a simplification. Furthermore, this neglects the single-vendor shop advantage when it comes to Field Service. And finally, does anyone know how long software updates for foreign controllers lag behind the releases from DEC? This last may only be meaningful for VMS; I don't know what AT&T's and Berkeley's policies are on this. You also don't say how much service will be done by locally trained techs and how much you will rely on DEC Field Service. How far away is the SI office? Is there really no cost difference? The monthly maintenance on the RA81's is amazingly low. Be sure to take that into account. For my setup, a number of large page-fault bound jobs at an otherwise all-DEC VMS site, the UDA50/RA81 combo was the clear choice. However, anyone who answers your query in 25 words or less is not answering it for you, they are answering it for them. Stew Rubenstein Harvard Chemistry
info-vax@ucbvax.ARPA (11/13/84)
From: engvax!KVC@cit-vax I have to say I agree completely with Richard Garland. The DSA disks on the UDA-50 are a low-end alternative to what DEC really wants to provide. The hypothetical 785 that was posed in an earlier message could be outfitted with a UDA and DSA disks now and later upgraded quite easily to a CI based system. I would porobably buy the CI now, however, and skip the UDA-50. I've watched system loads grow too many times in the past to underestimate the usefulness of the cluster concept. I also agree that VMS manages the UNIBUS adequately enough for a low-end system to survive with a UDA. I would argue, however, that someone trying to get the best performance possible out of the 785 should be concerned about the growth of the system in the future. The CI gives you an easy upgrade by allowing you to add more processors as you need them. Of course, the system posed hypothetically is going to run UNIX, in which case you have no cluster alternative. I think with UNIX I would stick to the MASSBUS since UNIX tends to favor proven hardware. My system right now has MASSBUS Eagles on an SI 9900 controller and an RP07. All very fast MASSBUS disks. I find, however, that my system is completely out of CPU power, while only using something like 10% of it's disk bandwidth. I would love to be able to plug in another VAX and split my user load without having to purchase any more disks controllers. (To be fair, a cluster WOULD allow me to use MASSBUS disks on any system in the cluster from any processor in the cluster via the MSCP server, but I know I would have to buy at least a system disk dedicated each additional CPU if I did not have an HSC). One final note: for a very thorough discussion of UDA, HSC, and MASSBUS performance, take a look at the article: Disk System Lantency in VAX/VMS by Richard Wrenn and Mark Freeman in the Spring '84 proceedings of the Digital Computer Users Society (DECUS). Although the HSC and cluster performance would be meaningless to a UNIX shop, you should take a look at the UDA vs MASSBUS figures. This is probably the best discussion of disk performance I have seen so far. /Kevin Carosso engvax!kvc @ CIT-VAX.ARPA Hughes Aircraft Co,
info-vax@ucbvax.ARPA (11/13/84)
From: Mike O'Brien <obrien@CSNET-SH.ARPA> Well, I'd probably scream a lot if given a choice between a single UBA with DEUNA and UDA-50, and an SI 9900. There are folks out there who are happy with SI controllers, and folks who aren't. On the other hand, there seem to be few folks out there who aren't happy with Emulex controllers. So, a) I'd plump for an Emulex. If I couldn't get that, on to plan b) get the SI. The contention between the UDA-50 and the DEUNA could get quite fierce, I think. Also, the DEUNA eats so much power that I'd be trying to keep as much stuff out of the UNIBUS cabinet as I could.
info-vax@ucbvax.ARPA (11/15/84)
From: medin@ucbarpa.BERKELEY (Milo Medin) In that case I'd get the Eagles, with no reservations. They are better drives. Milo
info-vax@ucbvax.ARPA (11/16/84)
From: medin@ucbarpa.BERKELEY (Milo Medin) I disagree. Massbus disks dont have to contend with tty's and ethernet and printer interfaces. I have seen some terribly performing systems that were unibus limited. When I added a 2nd unibus, and put the disks on it alone, performance improved significantly. Its quite marked. Average throughput on a sole unibus may be about the same, but thats a lot of trouble to go to for no reason, the massbus is there, and while it may be dumb, its a lot faster. Maybe ron is right, that they dont think they can compete in the Massbuss market. But thats not like them to surrender so easily... Milo
info-vax@ucbvax.ARPA (11/16/84)
From: hplabs!hao!pag@BERKELEY (Peter Gross) Mike O'Brien mentioned that you wouldn't want to share a UDA50 with a DEUNA cause the DEUNA is a power hog. Well, friends, check out the UDA50 -- it consumes 13.5 amps at +5v -- ouch! (Haven't seen the specs for DEUNA). --peter gross
info-vax@ucbvax.ARPA (11/17/84)
From: William LeFebvre <phil@rice.ARPA> A UDA50 consumes 12.75 amps, not 13.5. But a DEUNA consumes a whopping 15 amps! Double ouch! Both boards are power hogs. Unless you want very few terminals on your system, you cannot put a DEUNA and a UDA50 in your VAX without a second unibus. If you want a UDA50 and an ethernet interface on the same machine, you would be better off getting an Interlan board. William LeFebvre Department of Computer Science Rice University <phil@Rice.arpa>
info-vax@ucbvax.ARPA (11/17/84)
From: Richard Garland <OC.GARLAND%CU20B@COLUMBIA.ARPA> DEUNA, UDA50 and second Unibus: No you don't need a second Unibus! One BA11 box can supply 64 amps (50 amps on old models). You can string 2 BA11 boxes on the Unibus easily. That's plenty of room for everything. Too many people think they need a second Unibus adatpter (DW780) whereas they could get by with a second BA11 box. Each BA11 box can accomodate 2 large backplanes and 1 small. Thats about 12 hex slots and 4 quad slots per box. Constraints are total DC power (32 amps each half box) bus loads (22 max on whole bus - each interface is usually 1) total Unibus length (50 feet I think) and most important - system performance. Placement of interfaces on the bus, and bus interrupt levels must also be considerred. We have a working system (I wouldn't say it flies - but it goes OK) with the following: 1 LP11 6 DZ11's (48 lines) 1 DUP-11 1 DMF-32 1 DMR-11 1 DEUNA 1 UDA50 1 FPS164 interface (large array processor)ans 1 Evans and Sutherland PS330 interface This stuff is spread over 2 BA11 boxes on 1 Unibus. The system runs VMS and runs OK as I said. I plan to consolodate the 6 DZ11's into 3 Able boards when we get the funds. At least DZ11's needn't be considered by new installations. The main thing is - Consider a second BA11 box before considerring a second Unibus adapter. DEC sales sometimes forgets this option. Rg -------
info-vax@ucbvax.ARPA (11/26/84)
From: dual!unisoft!kridle@BERKELEY