From: Aaron J. Grier Date: 23:38 on 16 Mar 2006 Subject: yay raid controller I've got the following in my vintage late 90s alpha 1000A: mlx0 at pci0 dev 12 function 0: Mylex RAID (v2 interface) mlx0: interrupting at dec_1000a irq 3 mlx0: DAC960P/PD, 3 channels, firmware 2.70-0-00, 32MB RAM AKA the DEC KZPSC. it's a three-channel hardware raid controller, and is wired into a nifty storageworks shelf in the alpha. I saw that it was supported under NetBSD and decided to actually use it. first mistake. switching from SRM to ARC to run the configuration program isn't completely godawful; I had a monitor and keyboard hooked up when I configured and initialized my starter 16GB RAID5. no big deal, it didn't take too long to initialize the array, and the machine wasn't being used for anything yet, so I could afford downtime. second mistake. as old hardware from my alma mater is wont to do, a largish (126GB is large to me) nstor disk tray made its way into my basement. "wouldn't it be great to connect this to my hardware RAID controller?" so I bought the requisite HD68 to HD68 (honda) connectors to wire it up. third mistake. by this point I had moved the 1000A off my desk and into my rack, and given it a proper serial console. I wired everything up, fired up ARC, and... hey... god how awful this is over a serial console. still, I trudged along. fourth mistake. after some missteps involving the internal serverworks shelf being wired to the same channel as one of the external connectors (even though the controller has three channels) and figuring out that the nstor box doesn't need two separate SCSI channels anyway since it only has seven devices, I got everything ironed out and prepared to configure another logical device... it has HD68 connectors but only scans IDs 0-6, I assume since there are only seven slots in a storageworks shelf. I've got seven drives in the array, numbered 0-5, and 8. I don't know why I didn't renumber the last one to 6, but it's not terribly important, since I couldn't create a logical disk over 32768MB in size. so I ended up creating three logical disks; two 32.7GB ones, and one leftover. I should have given up at this point, but I figured I would just concatenate the logical drives together and just deal. fifth mistake? it took FOUR HOURS to initialize the logical disks. four hours of downtime, since the tool is being run out of ARC. oh did I mention that the firmware is too old for NetBSD's mlxctl to use, and I have a sneaking suspicion that if I updated it, SRM probably wouldn't be able to boot off it? (I may still replace the firmware and see what happens.) meanwhile I've got a nice 64bit LSI 53c1010 controller here that I wish I would've used in the first place...
From: Chris Devers Date: 01:54 on 17 Mar 2006 Subject: Re: yay raid controller On Thu, 16 Mar 2006, Aaron J. Grier wrote: > meanwhile I've got a nice 64bit LSI 53c1010 controller here that > I wish I would've used in the first place... Sorry, I'm confused. So the hate here is "getting crufty, outmoded DIY kit to do what I want can be way too complicated, can take way too long to get working, and may not work as well as more modern off-the-shelf kit" ? It may not be in keeping with the spirit of the list, but maybe, just maybe, user misjudgement (on top of more user misjudgement) is a factor here. :-)
From: Peter da Silva Date: 03:10 on 17 Mar 2006 Subject: Re: yay raid controller On Mar 16, 2006, at 7:54 PM, Chris Devers wrote: > So the hate here is "getting crufty, outmoded DIY kit to do what > I want can be way too complicated, can take way too long to get > working, and may not work as well as more modern off-the-shelf kit" ? Well, and "DEC writing software for wide SCSI devices that freak out at targets over 7". Tru64 was doing that up until 5.0, they even made 14 drive shelves that split the drives into two banks to allow for it.
From: Aaron J. Grier Date: 07:59 on 17 Mar 2006 Subject: Re: yay raid controller On Thu, Mar 16, 2006 at 08:54:47PM -0500, Chris Devers wrote: > So the hate here is "getting crufty, outmoded DIY kit to do what I > want can be way too complicated, can take way too long to get working, > and may not work as well as more modern off-the-shelf kit" ? that's expected with castoffs like this, and isn't in and of itself hateful. what wasn't expected was the 32768MB logical device limit and the four hour parity initialization, which are both firmware limitations. firmware is just a fancy name for software that executes on a processor other than the main CPU(s). the software RAID option, necessary because toor knows at least one of these drives will eventually fail, has the hatable read/modify/write hole: http://blogs.sun.com/roller/page/bonwick?entry=raid_z . > It may not be in keeping with the spirit of the list, but maybe, just > maybe, user misjudgement (on top of more user misjudgement) is a > factor here. :-) since when did this list include a restriction that the source of software hate be non-self-inflicted? (:
Generated at 10:26 on 16 Apr 2008 by mariachi