The collected works of jan-jaap - Page 2

AFAIK they were unbundled because SGI had to pay royalties per shipped CD to Adobe and/or AT&T. I've got versions here on separate CD's for various flavours of IRIX and I know it was on the Varsity progam CD's at some point (Varsity Update for 5.3).

thermionic wrote:

Code: Select all

...
need to seek backward to offset 64 (currently at 266768).
Closing and reopening file
bootpclose: ctlr 0
tftpabort: TFTP send ERROR packet


I don't know exactly what's happening here, but I don't think that's good.

thermionic wrote: The TFTP setup was trivial.

It's called Trivial FTP for a reason

Getting RSH to work the way that inst wants it to though, is proving a little (a LOT) harder, I've got a nasty feeling that this will take a while.... FC6 just wasn't having any of it, FreeBSD6.1 also is proving difficult. I can RSH onto the server, but getting a remote command to work is proving harder. However from what I've seen of IRIX so far, its really not intended as a secured OS, its designed to be inside a secure environmet </flame>

Two things:
1) The installation process utilizes r-* commands (rexec, rsh, ...). The reason you're not getting an love from an FC6 box is that e.g. "rsh ls *.idb" gives a different output then what the IRIX miniroot expects (something with XPG4 compliant wildcard expansion, 'color'-ls etc. etc.). Network installations from another SGI are a breeze. I used to run a Linux based install server but gave up. I'm moving it to a Challenge S (which still NFS-mounts the media from the Linux box).

2) Security: yes, this install method is from the days that the internet was still a friendly place Look at it this way: you can install any SGI (including the oldest 4D that otherwise need to be booted from quarter inch tape) over the network from any other for any IRIX version (3.x ... 6.5.xx) without modifications to the install server. I'd say this install server usually lives in the sysadmins room on a private network segment.

You can improve security of your IRIX box by installing IPFilter ( http://www.sgi.com/products/evaluation/ ... er_3.4.27/ )

ipaddict wrote: FreeBSD ships with an OSF/1 ABI layer that is designed for this sort of thing, as well. BSD is also generally well supported on the Alpha.

FreeBSD is dropping the Alpha at the next major release.
Linux has it's own problems as well: ever since the change from XFree to Xorg, Debian has been broken on many popular ATI cards on the Alpha. It was fixed only recently, after being broken the best part of 2006. There just aren't that many people interested in Alpha anymore...

Out of curiosity: where does the presenter card go, knowing that the XZ graphics option is a double card? Is it a little daughtercard or something?

I recently realised you can't install a fast ethernet or FDDI board in an Indy XZ

Grummz wrote: Is it possible to run widescreen on a 1600sw with an Indigo2 Max Impact machine?

I ran my MaxImpact for a (short) while at 1600x1200. The thing is: it works, but you loose so much of the graphics board memory to the framebuffer that few remains for Z buffer etc. Essentially, you turn this former 3D workhorse into a glorified 2D framebuffer

I don't even know if you can get 1600x1024 out of MaxImpact. And then you'd have to convert the signal into something the 1600SW likes. I don't think this worked for Octane with Impact graphics, so I wouldn't expect it to work for an Indigo2.

A lot of LCD's do work on SGI's. I got a freebie 17" Iiyama 430 the other day and it turned out it works perfectly with an Indy (so an Indigo2 should be a problem either). And that's 1280x1024@60Hz, something the SGI's default to

Not in the UK, but I have two Indys that have to go (spring cleaning).

Good looking / R4400 / 128MB / 24bit gfx / 2GB disk
I can probably supply Indycam and/or mouse+keyb too.

frood wrote: I'm a bit worried about import duty

Did the British finally leave the EU ?

Anyway, success finding your Indy closer by.

My IP21 (Dual R8000 Onyx) is giving me grief

As I understand, SGI made two families of R8000 systems : IP26 (Power Indigo2) and IP21 (Power Challenge/Onyx), and both are supported up to IRIX 6.5.22

Except 6.5.22, and probably 6.5.21 are broken on IP26 . Maybe even as far back as 6.5.18. I have reason to believe IP21 has similar problems .

I've had 6.5.14 running on it at some point and it worked. My IP26 runs IRIX 6.2 and is not affected.

So, my questions to those with R8000 systems: what OS are you running on them? Which is the last unaffected IRIX release? Is this issue fixed by a patch (that I would have to install when I install the OS)? Or maybe it's possible to run 6.5.22, as long as I use rqs binaries from an earlier, unaffected release?

I gave it another go with 6.5.22, and this time I included all relevant patches in the netinstall. That includes the rqs updates in patch 7009:

Oops!

And this time it actually worked

Oh, the PowerIndigo2 (IP26) stays at IRIX 6.2. I want 6.5.22 on the PowerOnyx (IP21) because I mostly run it with a quad R10k and I don't want to have issues with the potentially different IO4/FDDI etc. flash versions required/expected by different OS revisions.

Thanks all!

DDT wrote: I'll give it a try on the IP26

The crucial part is that the patch is installed before the rqsall run at the end of the overlay installation. That way you're sure to use the rqs from the patch, not from the 6.5.22 overlays. Hence the network installation (install servers rule )

Also, I've tested this on IP21, not IP26 -- if you nuke it you're on your own.

You should talk to 'loonvf', he's working on something similar.

nekonoko wrote: Since you have basically a bare motherboard with no fans (looking at your output), you'll need to disable environmental monitoring in order to power it up.

We tried that, but it doesn't even fire up the power supply.

We also tried the inverse: pull all memory, most fans, switches and leds from my Fuel and power it up from the L1. It does power up (and power down after 30 seconds due to the expected env. issues). At that point, the only difference between my Fuel and Francois' motherboard was that mine still had the graphics installed.

It's the "I2C: Error: <something>" message that's the problem, I think. And the question is: is it due to the missing grahics, or due to a broken motherboard?

Here's a picture of the guts of my Fuel (600MHz V12, first gen. power supply)

Boy, you're sure having fun with your little project, aren't you?

Added picture. System now also has BVO (Broadcast Video Option for VGX) so it has a whole lot of BNCs on the front panel.

pentium wrote: I thought it might be nice to finally upgrade from my tangle of 10Mbit BNC ThinNet, 10Mbit CAT 5e, 100Mbit CAT 5e, AppleTalk and Token Ring to just Token ring (so I can network my old AIX 1.3 box) and Fiber.

So, essentially you want to run regular ethernet using fiber as a medium instead of e.g. cat5 cable, using transceivers or SX cards. Usually people do that when they need to cross long distances (> 100m) but I understand the coolness factor.

pentium wrote: The only items I currently have that are related to fiber are two FORE ATM cards for the Crimson and Onyx and a card installed in my Indigo.

I hope you're not mixing up ethernet over fiber (using transceivers), FDDI and ATM. They all use more or less the same medium (fiber) but the protocol layer is of course totally incompatible so you can't mix them.

pentium wrote: What do you think?

I agree with you that it is a good idea to standardize on as little network media as possible. Personally, I'm using cat 5e for ethernet wiring (with AUI-RJ45 media converters where needed), and FDDI for systems for which fast ethernet adapters don't exist (SGI '4D'), are of poor quality (Onyx), or hard to find (Indigo/Indy/Indigo2).

You can buy a fan-less 16port gigabit ethernet switch, that consumes only 8W, fairly cheap, but an equivalent 1000base SX switch will be loud, big and power hungry. This becomes annoying if you run it 24/7. There is a reason these things don't have a lot of value in the second hand market

UrbanHero wrote: I mean....just look at all those blinking lights!

Get yourself a PowerSeries VGX and run it with the front panel open -- definitely up there with the 1960's James Bond movies

nvukovlj wrote: I believe that jan-jaap mentioned at some point that he archived off the Developer Toolbox site before it went down.

Affirmative

pentium wrote: Okay then, what specifically should I be looking for?

1) Read The Practical Guide to FDDI . All of it.

2) Search eBay for a concentrator like the Cisco WS-C1100 or WSC-1400 mentioned.

Be sure to buy one with the FDDI boards installed and not an empty chassis. Buy one with optical (MIC) connectors, if it looks like RJ45 it's CDDI (the copper flavour of FDDI) and it won't work with your SGI's.

With a little patience you should be able to pick one up for $25. Someone offered me three loaded ones for $50 some time ago but shipping was killer.

Oh, and those Cisco's don't have an ethernet port. But a Linux PC with an ethernet and an FDDI card makes an excellent router. At least, that's how I did it.

pentium wrote: If memory serves correct:
-Ethernet over fiber is really no different than regular ethernet.

Correct. Same protocol, different media layer. Allows you to cover bigger distances. Also referred to as 10base FL or 100/1000 base FX.

-FDDI works in the same way token ring does.

FDDI allows you to build ring topologies, like token ring. But at the protocol level they're totally different, so an optical transceiver won't allow you to mix token ring and FDDI. The downside of a ring topology is that you don't have a ring unless all systems are running. A concentrator allows you to treat FDDI like a star topology which is more convenient. That's all. Technically, you don't even need one.

-ATM will make your head explode if you fail to fail to grasp the concept of packet switching.

I have only one machine with an ATM interface, so I can't comment. But somehow I expect the details of the implementation (packet switching) to be hidden by the tcp/ip stack.

That being said, this would mean that I would require at the least a conversion from the FDDI protocol to the Ethernet over fiber protocol. If I wanted to use my two ATM cards in the Onyx and Crimson, I would then have to convert both ATM and FDDI to Ethernet over fiber.

You're getting it.

Then we got to remember that the FDDI and ATM converters as well as the Cisco gear will give off quite a bit of heat.

Don't forget the noise.

Why does networking have to suck so much?

No, you called this on yourself. You're enjoying the pain. Admit it

Schematically, this is what my network looks like:

To me, network infrastructure is not a goal in itself, it is there to support something else. So, I simply use what everybody else is using. It has to be cheap (to buy, but also to operate), and as simple as possible. I use regular 10/100/1000base TX copper wiring where possible. Convert everything that has a different media interface (AUI etc.) to RJ45 right away. Use consumer grade 1000baseTX switches. You can find fanless 16port switches that consume one a couple of watts.

Only where fast ethernet is not available do I use FDDI. Nothing else, no ATM, tcp/ip over fibrechannel etc. As you have found out by now, each new flavour requires a router to connect it to the rest of the world, and adds a new subnet to your local network which is a pain to administrate. As you can see, I have three subnets to administer, which is two too many as far as I'm concerned.

My FDDI concentrator is an IBM 8244, which is a 12-port device in 1U 19" form factor. It is fairly quiet and not too power hungry, as far as FDDI concentrators go.

I've got some pictures of (SGI) FDDI cards and the concentrator here: http://www.vdheijden-messerli.net/sgist ... 2.27-fddi/

If I were you, I would keep things as simple as possible for now. Establish a working baseline. Have you even tried to connect two systems via FDDI? Just a straight connection, and make them talk to each other?

That's about what I know of FDDI. One last thing: some of my SGI systems are dual homed (the Crimson and Onyx in this picture). In normal operation, they are configured to have their ethernet interface(s) disabled and use only FDDI. But the cable is there, so I can netboot them and perform an installation from the install server. Actually, if everything is configured properly, it only downloads the miniroot over ethernet. The actual software installation goes via FDDI. Makes a big difference if you want to install IRIX 6.5.22 + compilers + software on an Onyx1

The Keeper wrote: If you really want to get into fiber, and want to do something useful with it, forget FDDI and start looking at Fibre Channel.

Apples and oranges. Or maybe steam engines and diesel

For the first couple of generations of SGI's you have two choices, 10mb/s ethernet and 100mb/s FDDI. And maybe ATM, but certainly not fibre channel.

Later generations of SGI's might take fibre channel, but isn't it true that only the Prisa adapters for Onyx do tcp/ip over fibre channel (which was the whole aim in this case)? These same systems all have 100mb/s ethernet so FDDI is irrelevant for them, but for most of them gig ethernet is also available. Still, I admit that gig ethernet over fibrechannel in an Onyx1 has a certain geek-appeal

If you want to improve your resume, everything is better than FDDI, which is just another niche technology that went out of fashion a decade ago.

I have no experience with DEC or Cisco concentrators, but my IBM has a built in configuration menu on a serial console so I don't need anything special for it. I guess the default settings were sensible because I never had to change anything to make it work. Most systems (Indigo/Indy/Indigo2/Onyx/PC) were 'plug and play'. The only exception was the VME FDDI card for 4D systems. It's an Interphase 4211 V/FDDI board, which was modified by SGI, and they changed the fiber connectors around so I wired it up incorrectly. The fact that one of my V/FDDI boards was dead didn't make that one any easier to diagnose.

The Keeper wrote: I guess that's enough FC evangelizing for now.

Nah, you're the resident fibre channel guru, we wouldn't expect anything else

The Keeper wrote: [...] the poster of this thread wants to use orange wire. Sometimes more thought has to be given to enterprise-level technology than just colors, and the question "what do you want to do with it" has to be answered.

Hehe, somehow 'pentium's thread always get very long. Back to page 1:

pentium wrote: I thought it might be nice to finally upgrade from my tangle of 10Mbit BNC ThinNet, 10Mbit CAT 5e, 100Mbit CAT 5e, AppleTalk and Token Ring to just Token ring (so I can network my old AIX 1.3 box) and Fiber.

I think he wants tcp/ip networking, not a SAN. I wouldn't mind a SAN though Maybe once my KVM project is finished ...

The Keeper wrote:

deBug wrote: Only one of the boards has the optical adapter.

What you'd be looking for to complete the rest would either be called a "GLM", or Gigabit Loadable Module, or more likely at this point, Emulex LP6000 or LP7000 PCI cards with GLMs attached to them. Just yank the GLM off the LPx000 and drop it onto your cards. If you need more ports, that is.

What about bulkheads, do you have those? Or do you want to run fiber through a hole in the front panel of the Onyx straight to the card?

For me, that was a reason not to bid on that auction . . .

pentium wrote: Here is the plan of doom.

Did you realize you need a motherboard with a dozen PCI slots in the server?

If the L1 has an issue with the speed of the fan, it will shut down the system about 60sec after you power it on.

Here are 'normal' values for a Fuel:

I think the one you are replacing is the 'exhaust' fan, in that case 1200RPM should be OK.

I believe the more annoying fans of the fuel are the (small!) XIO fans in front of the graphics. Not only are they relatively loud, but their speed (and thus the pitch of the whining) goes up and down noticeably when you're using the system. The power supply fan is also rather loud.

SAQ wrote: Anyone here know why SGI tends to underdrive their processors (R10000 at 195MHz rather then the max rating of 200MHz

Dunno, but even at 195MHz they weren't very reliable when they just came out.
I had an account on a (then) brand new PowerChallenge (simple user) with ~ 12 x R10k's and during the first year the system was down quite often because a CPU had failed.

SAQ wrote: and I've seen a 4D/35 with a full 40MHz-rated R3000)?

4D/35's were dangerously close to the limit, thermally speaking. Maybe an underclocked 40MHz chip ran a little cooler? Maybe they didn't have any 36MHz chips left at some point? Who knows ...

jc179 wrote: The SI is actually really hot to the touch

Solid Impact is *nothing*

Try MaxImpact, or better: MaxImpact with the TRAM option cards I guarantee you it puts out two or three times more heat than Solid Impact. It's the sort of thing that noticeably warms up the room it's in. The fan in the graphics card cage can keep up with that, plus whatever you have installed in the fourth slot. OK, barely keep up

Anyway, I wouldn't worry about an SI card.

pentium wrote: I can still vividly remember calling it "the spider program" while I really sould not read at the time I still knew where it was.

It's called "ant, the movie" and it part of the demos (at least in IRIX 5.3). I've got a rather crappy shot of it HERE .

SAQ wrote: (a) prevent people from monkeying with system IDs.

(b) "You want to change your O3400 config? That must be done by SGI service - we'll have them send you a quote. Have you ever considered a service contract?" .

(c) To stop 3rd world countries with nuclear ambitions from evading the export restrictions by buying a couple of small systems and merging them into a large configuration. These days, with PC clusters, this argument has become irrelevant.

Hmmm, that error message doesn't sound good.

Have you tried to remove all memory except the first bank? Then, if the error stays, you install the SIMMs of the second bank in the first bank. I've seen really weird errors from systems with bad memory.

But if it's not that, it's probably the CPU board. I've got a spare one I can let go, but it's the 100MHz R4000 version. Plus I'm in Europe.

Here's some relevant documentation:


The bitter part is that the Crimson isn't even an MP system. But it interfaces with the MP backplane of the PowerSeries

ozpass wrote: Hmm. That's interesting.

In mine, banks 1 and 2 of each group are populated, so each group has 4 modules installed.

If I've read the instructions correctly then I should have two fully populated groups rather than having the modules spread over the four groups.....

Not quite.

One bank of memory is made up of 8 SIMMs, spread out over the four groups.
Memory is added one bank at a time, so either 16MB (8x2MB) or 64MB (8x8MB).

So your memory layout sounds correct to me.

ozpass wrote: P.S. Are these really "just" 80ns 72-pin ECC SIMMS? I was expecting something far more exotic.

Nothing special going on here. The memory chips are the same you find in contemporary PC's, only the form factor is a little different.

Arrays of zero length are a GNU extension: http://gcc.gnu.org/onlinedocs/gcc-4.3.0 ... ength.html

Usually, arrays of variable length (also GNU) are much more annoying to recode.