phil@amd.UUCP (Phil Ngai) (09/17/84)
> what are the pros and cons of upgrading the 9400 unibus controller > versus changing over to a CMI adapter (prices are roughly the same)? > > Al Conrad This isn't a direct answer to your question but I thought you should know. I've worked with an SI9900/9751 (CMI) and Emulex SC750/eagle and thought the SC750 was much more trouble free from both the installation standpoint and the reliability standpoint. Also the Berkeley paper on 4.2bsd filesystem throughput seemed to give the lead to the SC750/eagle combination. I am told you need a special driver for the 9900 if you want to do bad block handling (a must in my opinion) because more blocks are in the pipeline or something like that. SI sent a driver which didn't compile due to 2 missing constant definitions. I grepped for the two words in all the files they sent including their .h files and didn't find them. So I had to figure out what they were and invent them. The SI system has also had a couple of crashes. Field service of course says they don't understand Unix and are not especially helpful although they do try. A system administrator at Berkeley I talked to had some really nasty things to say about SI. Our emulex/eagle came up with no problems and has run flawlessly. I would take it for granted except I've worked with the SI system. It's too bad because SI seems to try really hard but in my experience their 9900 disk controller is to be avoided if possible. (this article is personal opinion and not necessarily endorsed by AMD) -- Legalized gambling is a tax on the stupid. Phil Ngai (408) 982-6554 UUCPnet: {ucbvax,decwrl,ihnp4,allegra,intelca}!amd!phil ARPAnet: amd!phil@decwrl.ARPA
irwin@uiucdcs.UUCP (09/22/84)
We are running 9900s on nine machines, 2-780s and 7-750s. The 780s have SBI interfaces on the mass buss, the 750s have CMI, so none are on the unibus. When we got the systems, we could not hack the cost of maintenance on all nine, so elected to handle the 750s ourselves. The 9900s on the 780s have been kept at current rev level, the 750s have not as a result of no maint. Our 780s are on DEC maint, but not the 750s. The firmware in the 9900s on the 780s is a 4 rom set (level 6.2 I think and handles bad block forwarding without special drivers) but we put in a patch for the bbf under 4.2 and have not changed it since the latest firmware was installed in the 9900s. I am told that level 7.0 is just around the corner. The firmware in the 9900s on our 750s is a three rom set, and we have had disk files damaged on those systems. We just went through a rather lengthy negotiation with SI to get the 750 disk sub systems on maint, so that we can get them up to current rev. I do not feel that the bug will remain after we have done so. We have been pointing the finger at SI because we were having problems with the 780s, they would go off into never-never land, no error messages and we could not determine what was causing it. SI helped us locate it, by studying the register dumps that we took. It turned out that we had some soft mem errors (which we knew about but ignored because they were correctable, were brand X, not DEC memory) and the memory controller was getting a second error while trying to correct the first, so "hung buss" by a confused mem controller. Now that we have cleaned that up, the 780s have not been causing problems. We got our 780s with 4 meg (256k boards) and later added an extra cabinet to one, moved the 4 meg to it and got a 8 meg mem for the vacated memory hole in the other. That one was the one giving us the most trouble and DEC mem tests on their micro diag floppy #3 for the 8 meg unit would not even report any errors. The #2 floppy which covers the 4 meg mem unit does a very complete test and points out any bad boards. If you have a choice of unibus or CMI or SBI, I would recommend avoiding the unibus controllers. We had them before changing over, you will get better through-put on the mass buss. We did not have problems with the unibus controllers (ours was Emulex) but a lot of swapping and so on can sure load down the unibus. We are running 96 RS-232 ports on each of the 780s on Emulex H2 unibus controllers and average 30 to 45 users logged in at any one time on both machines. I hope that this information is helpful. --al irwin--
kermit@BRL-VGR.ARPA (09/25/84)
From: Chuck Kennedy <kermit@BRL-VGR.ARPA>
Probably the greatest positive aspect of changing your controller
from the Unibus to a CMI adapter is the decreased load on your Unibus
If all you have on the Unibus is a few terminal multiplexors (DHs or what
have you), it might not be worth it. On the other hand, if you have
some heavy duty I/O hogs on the Unibus (such as some graphics devices
like Ikonases or Megateks), it might be worthwhile to consider changing
over to the CMI interface.
Also, just to provide a counterpoint to Phil Ngai's comments about the
9900 controllers, we've had very good luck with them here. In fact,
just a few months back I unpacked a couple of 9900 controllers and two
Eagles, plugged them in and they ran and are still running just great!
We have many older 9400 controllers as well. Our experience with the
9400 CMI adapters has been excellent. We have four CMI adapters on
our two 750s and they've run for two years now with no problembs after
a one day installation job.
I guess it just goes to show you can get good and bad opinions about
almost anything (just ask our maintenance guys about Emulex).
-Chuck Kennedy
U.S. Army Ballistic Research Lab
Arpanet: Kermit @ BRL
UUCP: ...!{decvax,cbosgd}!brl-bmd!kermit