mjb%hoosier.utah.edu@cs.utah.edu (Mark Bradakis) (12/19/89)
First off, thanks to those who offered their assistance with the SCSI on the PI. Still having problems, but maybe someday I'll figure it out. But now I have some questions about different disks on a different machine. We have a 4D/240gtx and we are interested in putting some SMD disks on it. Does anyone have any good or bad comments about doing this? What controllers, and how much do they cost? How is the IO throughput? I imagine that a fast machine like the 240gtx should really have no trouble with several gigabytes of disk hanging off the thing. Thanks for any info, mjb. mjb@hoosier.utah.edu "If I took a rollin' wheel, rolled it ten times 'round..."
markb@denali.sgi.com (Mark Bradley) (12/19/89)
In article <1989Dec18.113838.27875@hellgate.utah.edu>, mjb%hoosier.utah.edu@cs.utah.edu (Mark Bradakis) writes: > But now I have some questions about different disks on a different machine. > > We have a 4D/240gtx and we are interested in putting some SMD disks on it. > Does anyone have any good or bad comments about doing this? What controllers, > and how much do they cost? How is the IO throughput? I imagine that a fast > machine like the 240gtx should really have no trouble with several gigabytes > of disk hanging off the thing. > We sell a 4 port 6u controller. And we sell 1 GB SMD disks. You can expect 2+ MB/sec through EFS on sequential reads. In spite of some early bugs that needed to be fixed, this is presently a very stable product. I belive that the controller, disk in a rack, cabling and s/w is in the low $20K range today. Your sales person will have more accurate pricing information for you. As for buying and integrating your own, I don't recommend it, as you may or may not get the right rev. levels of f/w, h/w or whatever. And cable impedance is critical for these disk drives, as is the right grounding scheme, etc. I recently also heard that one of our customers mistakenly plugged a drive into the wrong voltage (easy to do) and blew up the power supply for the drive. But, that's my opinion (and in this case, maybe that of my employer...) markb -- Mark Bradley "Faster, faster, until the thrill of I/O Subsystems speed overcomes the fear of death." Silicon Graphics Computer Systems Mountain View, CA 94039-7311 ---Hunter S. Thompson ******************************************************************************** * Disclaimer: Anything I say is my opinion. If someone else wants to use it, * * it will cost... * ********************************************************************************
rayan@cs.toronto.edu (Rayan Zachariassen) (12/19/89)
markb@denali.sgi.com (Mark Bradley) writes: >As for buying and integrating your own, I don't recommend it, as you may or >may not get the right rev. levels of f/w, h/w or whatever. And cable impedance >is critical for these disk drives, as is the right grounding scheme, etc. It really isn't that bad... it is hard to go wrong in hooking up an SMD drive (although when you get 220V out of 110V sockets...). The more serious problem is getting disks that work with your controller that work with your vendor-supplied or home-grown driver. We are not enamored by the default controller on SGIs, nor the preferred disks, and this combination is indeed sensitive to rev. levels just now being fielded. I would recommend using SGI's supported configuration (from SGI if you value your grey hairs, I gather) unless you have very specific needs or experiences that make you feel it is worth the hassle of listening to SGI people saying "told ya'". I do know for a fact it is possible to do everything homegrown, both an SGI-subsystem-clone configuration [which is what you're probably interested in] and a specific controller/disk combo we like to use around here on heavy timesharing Sun configurations (Ciprico Rimfire/Fujis). >I recently also heard that one of our customers mistakenly plugged a drive >into the wrong voltage (easy to do) and blew up the power supply for the >drive. >But, that's my opinion (and in this case, maybe that of my employer...) I hope so (or if its SGI speaking, hope not). Our machine was inches from the machine referred to here, and the mildest retort to that statement would label it "revisionist history". The victims may comment themselves. However, after the event we were all wondering what the point of having fool-proof socketing standards -- incompatible sockets for different voltages -- is, if they aren't applied (no, the customer didn't supply the metaphoric fool). ps: frankly, at this point, all SMD subsystems are bad relative to the speed of the rest of the system, its just that some are worse than others for various kinds of activities. If we can afford IPI subsystems or some other faster I/O technology, we'll be buying that for our machines asap when the money can be found. rayan
andrew@alice.UUCP (Andrew Hume) (12/20/89)
In article <1989Dec18.113838.27875@hellgate.utah.edu>, mjb%hoosier.utah.edu@cs.utah.edu (Mark Bradakis) writes: > We have a 4D/240gtx and we are interested in putting some SMD disks on it. > Does anyone have any good or bad comments about doing this? What controllers, > and how much do they cost? How is the IO throughput? I imagine that a fast > machine like the 240gtx should really have no trouble with several gigabytes > of disk hanging off the thing. we have a 4d/240 that came with one 1.2GB smd disk on a xylogics controller. we thought to upgrade the system ourselves by buying our own disks (i should point out at&t has serious volume discounts with imprimis.) so we bought the cables from sgi, the disks from imprimis (sabre) and installed them into the disk farm. before i got around to formatting them, we installed 3.2. so i tried to format the new three drives. one formatted okay, the others failed (too many succesive errors). (in passing, i scrogged several inodes on my system disk because fx causes the controller to reset and it looked like we lost an inode or two every time that happened. thereafter, i did this fx stuff standalone despite what the hotline says.) i mkfs'ed, fsck'ed and mounted filesystems on the new disk and very soon had a console full of hard ecc errors. having no fear and feeling no pain, i called the hotline. they very soon said that if i wanted to use third party stuff that was my problem. i said, well that's okay but at least you should tell me what firmware revision the controller, disks etc should be (after all, i had duplicated the hardware sgi sells). hotline agreed and called back sometime later. this time, hotline was apologetic and polite and said sgi doesn't do anything to the drives anyway and since we had a full maintenance contract, hotline would help. well, they still didn't have the revision numbers but i said i would try reseating all the cables etc and checking all the dip switches were the same etc. a few days later, no hotline callback so i tried imprimis. i got hold of their smd specialist and said i had problems with sabres on an sgi box. he said, 'hmmm', i heard pages flicking and then he said 'yeah, a lot of that starting around september'. well to cut an already long story short, this was a known problem with three causes: a tiny fraction was due to a bad prom on the sabres (long fixed), a bunch more problems were due to fx (these were supposedly fixed in 3.2) and the rest were due to controller problems. so i called the hotline and they asked 'well, how are things?'. i replied that i knew what the problem was and relayed the news from the imprimis guy. the hotline guy said he knew about all of that (sure! why didn't he tell e before?). he then said we needed firmware revision 2.6.2 or later on the xylogics controller. i had already pulle dthe board to find we had 2.6 and so we quickly agreed for a new board to be fedex'ed to us. i just formatted one of the new drives today. no problems (although it has not been soaked). i was not satisfied with how the hotline handled this but at least now my disks seem like they are working.