The collected works of SAQ - Page 16

Yep, pretty boring. The absolute worst has to be "Horse With No Name" - vapid lyrics and insipid music that seems to go on forever.



Ah, but there you have the power question (along with noise, space, and risk of failing mechanics). An "embedded platform" setup (Atom, MIPS, ARM) with sufficient expansion would be a lot easier to deal with.

The IP should be sent over with the bootp requests.

You've followed the SGI remote install guide to the letter, right?
http://techpubs.sgi.com/library/tpl/cgi ... /ch02.html

On a sad note it looks like SGI has broken or eliminated the online TechPubs for 6.3, 6.2 and 5.3. Are there any mirrors around?

Not positive, but that might fit a PM1 (100MHz) machine better than a PM2 machine.

It's a pre-ARCS machine, so any disk that will work in the other 4Ds will work in it.

One good (but slow) one that pops up (fairly) often is the Toshiba XM-4101, AKA the Sun SL-CD or SPARCstation internal CD-ROM. It's got the jumpers to select firmware - solder in switches and you have a rugged and almost-universal "old computer CD-ROM drive" I don't think it works for IBM RS/6000, but that's about it.

Do they make a hyperdimensional model for those tricky left/right things?

I thought you understood U.S. business types better than that. They do that and they'll have to deal with people asking for support and also totally annihilate the market of people eager to pay $1k+ for the software. Besides, giving away something cheap with no hope of big bucks later is hard to explain to your board when you're trying to justify your request for a raise.

That and they'd have to go through and re-figure all of the royalty payments due on the software from bits taken from other companies (or clear that it is truly 100% theirs), find an old disttape, post it, figure out how to generate licenses, etc. It would cost them several thousand dollars to get the program up and ready to go, methinks.

Trying to make a decision on a backup plan.

Currently have a dying NAS that is configured RAID0+1 for "pseudo-versioning" backups (not really versioning, but it will give me "file X as it existed at 21:00 on the last day in November" if I do something stupid). In concert with this there is the primary filesystem (RAID-1 mirrors) and a rotating offsite image backup.

Replacing the NAS, and I was considering initially going with RAID-6 (really the "2-disk redundant" Synology setup) so I can potentially recover from any two drive losses (rather than any 1-drive loss and some 2-drive losses with RAID 0+1), but then I started thinking. What is the likelihood of ever needing that level of redundancy for a "midlevel online" backup? Would I be better off doing RAID-5 (Synology "1-disk redundant)? The NAS (new one is Synology DS413) isn't in a area where it gets "bumped" physically, electrically or thermally - so the risk of loosing everything (i.e. all backups+ original data) is very minor.

I'm not expecting personalized consultation here, just interested in what others are doing. If you have a NAS in your backup plan what do you have it set to?

It can be part of a backup solution - in this case the part to deal with the days when something gets inadvertently deleted, overwritten or corrupted.

The offsite images are for the disaster recovery should we need it.

Yeah, tape's great for backups, but terabyte+ capacity drives are several thousand dollars so we're on hard drives for the high capacity stuff. This setup's going in a nonprofit that I do computer support for, and I don't want to spend my off time tape swapping or trying to teach someone how to do it - with a simple hard drive swap I can get one of the employees to manage the offsite backup.

Heck, I'm on rotating media for backups at home because swapping TZ87's gets real boring real fast.

For service storage I'd only consider mirrors or RAID-6. Service filesystems get too much of a beating, and the risks of a error during rebuild of a large RAID-5 are getting too high - and restoring from dumps is not fun. I'd even prefer RAID-6 to RAID 0+1 now for a 4+ drive setup, since RAID-6 allows any two drives to fail, whereas if a mirrored pair fails in RAID 0+1 it's restore time .

The limiter now is still platter speed. At 6Gbps with 4 channels you can throw a lot of drives on there before you start to notice performance loss, unless you're running stuff that fits in the disk caches Disk transfer rates are still in the 200MB/s rate for 15K drives so if you balance across all channels that's what, about 12 drives before you have a potential cut in performance under ideal situations?

If the machine turns on with green LED and doesn't do anything else (no switch to yellow LED) then it's something very early in the bootup process - bad processor module, very bad IP2(2,6,8)(memory controller or such), bad PSU (no +3.3V on faster-than-150MHz machines might do it), or on the Indigo2 possibly bad bank-1 memory (you do have a full matched quad, right?).

If you get terminal output then that will sometimes tell you right where to go.

I'm thinking I'll go with RAID-6 instead of 10 (even though it's slower) for one big reason:

On the last NAS we had a disk dropping off the array from time to time (not a specific disk, disk worked fine, disk listed as supported for the NAS, we're not sure what was up there). If a disk had dropped off and another failed then there's a 1/3 chance that we'll loose all the data on RAID10 (since we have 3 disks, 2 of which are mirrored and 1 not mirrored). If we're on RAID-6, one disk dropps off and one fails then we can still rebuild.

With a Synology (as opposed to Western Digital) NAS the chances that it will work reliably are much higher, but we can take the performance hit for the improved recovery.

If there's any chance of parts swapping at any point it also will pay to make sure that all components are compatible.

R4k machines (big aluminum heatsink, connector mounted on board, no fan on processor) need an IP22 mainboard.
R8k machines (smaller heatsink, connector mounted on a flex lead) need an IP26 board. Rare and a bit delicate.
R10k machines (large shiny aluminum heatsink, fan inset) need an IP28.

IMPACT graphics (large heatsinks on the board) need an IMPACT riser (white connectors) and IMPACT power supply (flat blue pigtail + one twisted pair of wires connected to the graphics riser). Newport and Express graphics can run in either.
IMPACT doesn't work in R8k.

It'll work for OpenVMS AXP, not sure about HP-UX.

You can get Visualize-FX cards pretty cheap and they'll work in most 64-bit PCI capable 9ks.

HP-UX doesn't make much use of graphics anyway, and most of that is standard X calls, so a remote X-server is eminently possible on these boxes.

This part NAS is just for someone who accidentally deletes something or corrupts something. The "real" backups are rotated offsite for disaster recovery. They're not as redundant and they're harder to get at, hence this midlevel layer. There is always a recent copy of the current filesystem offsite, so if the building burns we can be back up the same day we get new equipment. However, people will be out of luck if they also ask for the e-mail they accidentally deleted a week before to be restored.

Yeah, I gave them a scenario for tape, but it would need to be LTO-5 or LTO-6 for single-tape backups, and between the cost for equipment and the discussion of tape rotation strategies interest waned significantly. I set them up with something that will get rotated out regularly for offsite (external hard drives) and will, via the NAS, allow for speedy recovery of userbloopers. High on my list is the fact that it's also easy enough for the people who work there to take care of it once I set it up. I don't begrudge them the time it takes to fix things, but given the choice between coming home after work vs. running over to their spot to fiddle with something on a daily basis...

NVRAM is not an issue with Indigo2 (Indigo1 doesn't like bad batteries at all, though). The MAC/serial number is stored in separate chip from the Dallas module.

Since none of the things you've done have worked it's looking more and more like the processor module. Fortunately new PMs for R4k machines are still inexpensive. Before junking it I'd check voltage levels just to make sure, and also ensure that the processor module is firmly screwed into the auxiliary standoffs (the standoffs that connect to the bit of PCB protruding from the heatsink) if it's a greater-than-150MHz machine (the heatsinks provide the +3.3V needed by these processors).

You might want to specify if you'll take any, just the "E100 and clones" (3com EISA) or you want the Phobos GIO card.

When Indigo Magic was made the base SGI machines were 8-bit (Indigo LG(x) and Indy Newport-8). It's designed to look OK on an 8-bit display.

If you put a 24-bit image (a gradient is a good test) on the root window as wallpaper it should look 24-bit.

You're forgetting the most important part - get in there and check it out "by hand" regularly (every week or two).

Odd that the seller took off the extensions, since they're a big help in fitting boards into the double-deep IRIS.

68k-IRIS never shipped with SCSI, only ST506 (with RLL I think) style and ESDI.

SGIs are general purpose computers, so they'd be capable
... but ...

the graphics hardware is definitely not a GPGPU. If I recall, ICE was able to be accessed by some software (cf the Photoshop accelerator), but I don't think SGI ever publicly released the docs on how to do this. Seems there is/was an effort to get some support in Linux ( http://www.linux-mips.org/~glaurung/ ), but I'm not sure where it is now.

On the other machines the graphics are pretty much one-way devices - shovel data in one way and get a display signal out the other. If you really work at it you might be able to get the Command Processor in InfiniteReality to do some stuff for you, but then you're just using a ARM slave processor, which isn't too special. I don't think the computer end can access the GEs directly in any of the big SGI graphics.

Danger, Will Robinson.

You'll wind up keeping both.



You must eat very deep-dish pizzas.

Why stop at 20? O3k will go up to 1024! One of Nekochan's Aussie members had a 128 processor unit. With these you don't need to worry about message passing or any of that, just make sure your code is sufficiently threaded.

Depending on what you want to do with it. There is one catch with XS graphics - if you don't have the Z-buffer then it is not emulated in software (I think this is the only instance were a GL feature is not emulated if it isn't present in hardware). Look for XS-24Z if you need to do 3-D (or stick with LG).

You may need to change the brick type of the Tezro - I'm not sure if the PROM checks the type to determine if it will link I know you can't use a router with "Tezro" bricks.

You could also NUMAlink two Onyx350s.

Is it a "tophat" type similar to the other Dallases and MOSTEK units? If so, then there's probably a crystal and battery in the tophat and a little creative excavating will allow you to get at one of the two batteries (and since it claims to automatically switch it should ignore the dead batt).

I'd look at either of the two "end" sides for hidden pins. X-ray would work great (are you friends with a dentist?), but you might be able to track it magnetically too (or just take a guess based on how the potting compound goes).

IP is often used as an easy way to get a head on an O3x series (similar to CHALLENGE EXtreme). IR is probably the easy way to get big SGI graphics performance on a server, and it offers pretty much everything that SGI ever offered graphics-wise. IP looks to be a bit more work to get running.

If you really want a challenge you can always try UltimateVision.

SGI startup sequence starts with a green LED until the processor sets it to yellow (at the start of the diags). A immediate-and-always-green LED is a Bad Sign. Try reseating the processor module and swapping RAM around to get different sticks in bank 0.

Make sure it's non-Indy 5.1.1. 4.0.5 is better for low-memory systems, or perhaps even a base install of 5.3, turn off all the gegaws and install the X11 port of 4Sight's environment from 5.1.1

5.1 had terrible memory leaks and I can't recall how much 5.1.1 fixed.

Note that it's BYTE not bit. ODYSSEY doesn't do 8-bit.

Why do parity now? ECC does everything parity does and more, which is why you only find parity now in caches. DRAM is cheap enough you'd be silly to go parity instead of nothing or ECC.



Longer lifetime if you're writing a lot?

Catching up on some old mail. Looks like VAXMAN finally had to retire FAFNER:: from daily service.

Fafner was (and is) a VAX 7820, one of the last publicly-accessible big SMP VAXes.

More here: http://northtech.us/content/20130123/end-era-rip-fafner

I think 5.2 is new enough to require the password for single-user mode.

You can get a Linux distro that has EFS read-only support and SGI disklabel support, grab the password file, and run it through John. For 5.2 it's a basic passwd, so should be fast on any modern h/w.

Or then there's always the "I don't have anything that reads that" approach for UNIX machines. Open the disk on a hex editor, search for the string ":0:0:", then either snag the password to run through John or replace it with a known password (you need to have a compatible version, so don't use the password from your new UNIX box using Blowfish or whatever.

Can't recall specifics, but I'm looking at an old Apple CD-ROM here that I marked "No VAX" on. Can't say why, but it must not have worked. I use an XM-4101 for DASD VAX installs.

My concerns are time, security (partially solvable by encryption) and what happens if they decide to pull the plug tomorrow. Contrariwise, how do you verify their backups (or even for that matter verify your backup in a timely manner)?

I go SCA and an adapter pretty much all the time now. Used SCA drives are very cheap and pretty new, and you can get the "universal" 50/68 -> SCA adapters that will allow you to move your drives to almost any machine. Older versions of IRIX won't always read newer versions of XFS, but you can use EFS or even tar to the device for a fast file transfer. Also works for IRIX installs.

Your cables are more likely to fail than the adapter.

Oftentimes when there's something really odd like that it's because something is certified to run on a certain specific piece of hardware and it won't run on anything else (think O2+ and medical imaging machines, and there's something that runs on AlphaStation 21164s as well). Unlikely, but also possible, are certain control things that require certain I/O busses (think PDP-11 and VAX for industrial process control w/ custom QBus I/O or custom VME h/w on something).

Not that all instances of machine X are worth that, the dealers just have guaranteed working models that they can get to someone when they need it and with support for that much (and then there are the other people who either are delusional and hope to get lucky selling something "as-is" and with no support for the same price because they think that's what it's worth).

Can't tell you what the deal is with the B2600, but off the cuff it looks like that came with FX5 graphics. At that time HP was trying to show that they could equal SGI in raw performance (though they didn't have all the software and library backends that SGI offered), so it's very possible that some medical (or other big-money) device manufacturer used "HP's most affordable PA-RISC workstation" on their device.

If you're handy with a soldering iron you can wire in a VGA connector beside the EVC connector (or get the HP adapter, part #1253-0635). Pinout: http://www.hardwarebook.info/Enhanced_V ... _%28EVC%29

Personally, HP-UX is the sort of OS that doesn't seem to really care if it's on a local head or remote "vendor X" X server display (unlike IRIX, which really does like SGI graphics hardware), and there aren't very many killer apps that made use of the fancier graphics, so I usually just do TTY and network access.

If you haven't checked out OpenPA that's a good resource ( http://www.openpa.net/systems/hp-visual ... b2600.html ),