The collected works of jan-jaap - Page 16

vishnu wrote: Here's a totally off topic question, did Visual C++ exist on AlphaNT?

Yes, up to version 4.2, IIRC

miod wrote: Funny enough, on the identification sticker, some of them sport `0410' on the top right corner, while the others sport `0411', but there are no other visible differences.

Probably different production dates, week 10 resp. week 11 of 2004.

If the scrap value of an Onyx2 is > $300 then there must be more gold in my computer systems than in my wife's jewelry box

BetXen wrote: Did you also see that the processor speed is 50MHz ? In the OS, hinv correctly reports 100MHz. Why ?

Before the R4000, the internal and external clock of the CPU was the same, e.g. a 33MHz R3000 runs internally at 33MHz and so does the mainboard (or at least the bits interfacing the CPU).

R4000 (and contemporary Intel CPUs) introduced clock doubling : the internal frequency is a multiple of the external bus frequency.

In case of the first R4000 SGIs, the tech guys referred to the CPU by it's external frequency (50MHz), the marketing folks went with the highest number (100MHz). In the end, everybody went with the highest number, but the PROM still mentions the external frequency. Nothing to worry about.

These days with multi-GHz CPUs, the multiplier is of course much higher than 2 or 3x.


PS: nice looking Indigo!

jodys wrote: Then again, I can't put a R5K in an Indigo which means the Indigo doesn't run mips4, limiting the nekoware I can run. Thus, perhaps the Indigo should be a Indican't

That's hardly the fault of the poor Indigo.

Just install IRIX 5.3 or 6.2 and TGCware. Enjoy the increased speed and responsiveness of your Indigo-go-go

I paid €250 for my own Onyx2 deskside. At the time it was a dual R10K config with IR2E graphics, but with pristine skins. I sold a couple of Onyx2's a while ago for prices between €200 - €350, or traded them for something 'interesting', like a pair of Challenge L desksides. These were mostly 'rackmount deskside' 300MHz R12K systems with IR2E graphics.

guardian452 wrote: How to get a good price is (a) know where to look (industrial/university surplus auctions, etc) , or (b) know somebody who wants to get rid of one.

Don't forget (c) plain luck or (d) someone who tips you off. Thanks to option (d) I picked up a purple O2+ a couple of weeks ago for €25.

And that was only because I was generous (the money was going to someone's kid). The asking price was ..... <insert drum roll> ... 10 bucks.

VenomousPinecone wrote: As for the very fine black dust, it is probably just run of the mill toner.

No, I've seen that in Onyx2's etc as well. I believe that in a system with sufficient airflow the larger dust is simply blown out, and only the smallest particles remain.

You can clean it off with a clean paint brush.

These patches were superseded by others.

patchSG0003236 replaces patchSG0000956
patchSG0002605 replaces patchSG0001003
patchSG0001826 replaces patchSG0001015

Those are (were ?) available from Supportfolio

EDIT:

Which patches are needed for 5.3 XFS?

I assume the recommended/required patch set for IRIX 5.3 w./ XFS. There were two of them: 5.3xfs_Challenge_Onyx.tardist . and 5.3xfs_Indigo_Indigo2_Indy_ChallengeS.tardist .

PS: Yeah, that's my FTP server. Those patch sets used to be widely mirrored on anonymous FTP but they seem to have disappeared so this is my service to the public.

jpstewart wrote:

jan-jaap wrote: Now if only I could shut up that bloody Silkworm, it is absolutely intolerable....

What model of Silkworm? I recently acquired a pair of 32-port Silkworm 4100s (well, actually, the IBM-badged equivalent) that are intolerably noisy for about 2 minutes after powering them on. Then once their embedded operating system has booted (and the fans are under control of the environmental monitoring software) they quiet down to a much more reasonable level. Maybe a newer Silkworm (or even a firmware update for yours) would help your situation?

I'm happy to report that the HP SANswitch 2-16 (a.k.a. Silkworm 3800) has been replaced with a shiny new^H^H^H second hand EMC DS-5000B (a.k.a. Brocade 5000). The 5000 is the successor to the 4100. It's also noisy at start-up (though less than the 3800), and then settles down. It's still an 1U unit and those fans still run at ~ 7500RPM, but it's much much more agreeable.

Strangely, the airflow is back-to-front. Ideally I should mount it in the back of the rack, but that's not very convenient. Think I'll ignore that for now.

As a side effect of this noise canceling effort I now have 32*4Gb connectivity instead of 16*2Gb Took me a while to find one with 32 (licensed!) ports for an acceptable price, and on the right continent. As a bonus, it also come with Extended Fabric, Fabric Watch, Performance Monitor and Trunking licenses. That should keep me busy for a while

vishnu wrote: What kind of storage are you going to feed with that beast?

Uhm, I've got an IBM DS4300 SAN server (with an extra EXP710 disk tray) and an SGI TP9100 JBOD. Both have dual 2Gb attachments.

I'm gradually upgrading my infrastructure to 4Gb. Several of my SGIs have dual 4Gb cards already. I don't expect to upgrade to 'real' 4Gb FC SAN hardware anytime soon. All disks are 2Gb as well, for example so I'd have to throw out everything. A more likely upgrade path would be that my next file server will run an FC adapter in target mode. With a half decent RAID adapter and SSD caching that should be plenty fast. And I wouldn't need 28 spindles to get 8TB of storage capacity

jpstewart wrote:

jan-jaap wrote: Strangely, the airflow is back-to-front. Ideally I should mount it in the back of the rack, but that's not very convenient.

That seems to be somewhat common on switches. I guess in a data centre environment it makes sense to mount the switch at the back of the rack. Putting the switch ports at the back where the servers' ports are can simplify cabling --- no need to pull cables through to the front of the rack. I'll have to check the airflow on my 4100s now. I never thought about that until you mentioned it.

There's another possibility: the only place where this thing generates any significant amount of heat, is the cluster of GBICs at the front. Airflow front to back would draw the hot air through the chassis. Back to front probably results in lower chassis temperature.

vishnu wrote: And your main usage of that enterprise-class storage is for photos of the wife and kids, right?

When I got the DS4300 it had no disks, so I only had the controllers to mess with. I quickly figured out that they run vxWorks. I deal with vxWorks professionally more or less on a daily basis so that got my full attention. Now I have all the optional licensed features like flash copy, volume copy, FC/SATA mix etc etc.

Up until this moment this thing was nothing but yet another way to waste my free time. I got it in exchange for a charity donation so it didn't cost me much. Then I figured out this thing actually works with IRIX clients, unlike some other SAN hardware. It's an Engenio unit, also rebadged as the SGI TP9300. So I bought some disks for it. At some point I found a cheap EXP710 (extra disk tray). If I run into a cheap EXP810 I'll pick that up too -- it supports SATA disks, and if I hack the carrier firmware to accept generic SATA disks I can offload all my old 1TB disks into it.

And it's used to store funny cat pictures, what else?

hamei wrote: I don't suppose that the entire desktop hanging solid when an nfs share disconnects will ever get fixed now

My MacBook used to do that, but it's fixed in Mavericks (or the big cat release before that which I skipped). So it can be done

cesss wrote: Most single processor SGI workstations have the limit of 1GB RAM (except the Fuel with 4GB).

What (64 bit) single processor board do you think would need less modifications (mainly in number of chips and ASICs) in order to support a larger number of RAM GBs?

Least modifications would be a single CPU Octane2: zero

cesss wrote: Possible candidates I think would be the Crimson and the Indigo2, but I don't know how complex is their respective memory logic.

The concept of 8GB RAM in a system with a 100 or 150MHz CPU and a 10MB/s SCSI bus (or was it 5MB/s sync? can't be bothered to look it up atm.) is ... ahem .. interesting

cesss wrote: And now... do you think IRIX would be able to use 8GB RAM on a Crimson or Indigo2 if you had such a modified board?

No.

It's physically possible yo install 1.5GB of RAM in an Indigo2 R10000, but if you install more than 1GB it won't boot because something else is mapped into that region of the address space. I tried it myself and it doesn't work.

So, not only would you have to modify the hardware in an incompatible way (new IPxx number), you would have to change the OS as well. While you're at it, you might as well upgrade the SCSI bus, network interface and a few other things. This has been done already and it's called the Octane

Your best bet would probably be to butcher an Octane and mod it into the Indigo2 chassis. If you dare do this with a Crimson I'll come over and kill you myself

Another, more interesting angle would be to replace the RAM chips of O3K class DIMMs. They are mostly like normal DIMMs: DDR memory, SPD chip etc. but different form factor. Replace the RAM chips and reprogram the SPD to create 4GB RAM kits. Who wants 16GB in their Tezro? I do

Jack Luminous wrote: old swiss reel-to-reel machines

Studer/Revox? Kewl. I swear, one day I'll buy a Mk II B77

If nothing happens (no LEDs, fans etc), it's almost certainly the power supply.

This is not uncommon for Indigo2's, there's a guide out there how to recap the PSU. My success rate doing this is ~ 50%

foetz wrote:

vishnu wrote: Now hacking out the accursed variadic macros one by one

why? mipspro does support c99.
to be sure i just ran a quick test and used one from the gcc(!) examples - it worked

Certain headers (stdint.h, for example) are marked as 'C99-only', but only the MIPSpro C compiler supports C99. Try to include <stdint.h> in some C++ code and compiler with MIPSpro to see what I mean

GCC C++ allows C99 features in C++ code.

ClassicHasClass wrote: Another year, another machine room panorama.

http://www.floodgap.com/etc/machines.html

Re: the "ghetto halon system": get rid of it. If it's powder at least. That powder is a salt, if the moment ever comes and in a panic you empty that thing, you can write off all your hardware due to corrosion issues 6 months later. For an electronics fire CO2 is best. But big and heavy.

Same for the extinguishers many people have in their car, btw. If you car catches fire: let it burn. Assuming everybody got out, of course Insurance covers it. If you use a powder extinguisher, the insurance will say "thanks", repair your damage, but again 6 months later you will suffer a massive electronics failure, probably write of the car and it will *not* be covered. You did it yourself, after all.

Trippynet wrote: Oh! and I also reversed the flow of the fan on my system - default was to blow air OUT at the front, and I don't like fans at the front and back both trying to extract air

Not sure, but I maybe it was originally back to front because of the way the ASICs are placed on the graphics boards. Especially with IMPACT graphics, all the big ASICs, TRAMS etc are located near the front of the machine. So with air flow back to front you don't pull the hot air over the rest of the PCB, but it goes straight out instead.

I've got Brocade 5000 FC switches. They use ~ 50W (which isn't much for a 32 port 4Gb FC switch), and nearly all of it must be consumed by the SFPs. Which are at the front, obviously. So the airflow is back to front. If the hot air were guided out the back, you would need additional airflow to keep the temperature of the internals within bounds.

Oh, and I used to have a cat who loved to sleep in front of my MaxIMPACT 10K Indigo2. Didn't care about the noise, loved the hot air flow

I think a new thread is justified here

Subject: WTB: Octane2 2x600Mhz (or 2x400Mhz) R14K with V12 Pro

recondas wrote:

jan-jaap wrote: One of those plus a suitable monitor just might make a decent replacement for a bulky/expensive (HD)SDI monitor.

Nice find - compared to the name-brand-high-priced-spread they are indeed very affordable.

Well, I ordered one (not that one but another, with embedded audio support). I guess it's going to take three weeks to deliver and will arrive in a bubble wrap envelope Still, I'm completely in the dark about many aspects of the DMedia puzzle.

I mean, I've got the SGIs covered:
* Onyx2 with HD-GVO (DG5-TVO) which does 720p or 1080i (amongst others), DM2 (1080p), VBOB w./ DM5 (I hope it matches, borrowed from Octane2). This should give me HD-SDI video in/out at 1080p and HD-SDI GVO at 1080i. Oh, and I still have a DIVO/DVC card somewhere.
* Tezro with V12+DCD and DM3, VBOB (same as the one from Octane2). This should give me HD-SDI video in/out *and * GVO at 1080p (?? not sure about capabilities of DCD, DM5 combo).
* Octane2 with V12+DCD and DM2 for good measure.
* Enough FC interfaces in the systems and disk arrays to provide the required I/O bandwidth.

Now I want to *do* something with them Get some sort of video monitoring, and feed it it video data. This will have to use the SDI and HD-SDI BNC's from the DMedia gear. Let's see what I have already:
* a HD-TV with HDMI inputs (too far away and not portable)
* a mini-DV consumer cam, SD resolution, can play *and* record via 1394, s-video and composite.

Oh, something else: looking at the SDI doc: http://en.wikipedia.org/wiki/Serial_digital_interface I need 3G SDI to support 1080p (@50 or 60Hz). The el cheapo converters do that, but the 3G spec is from 2006 so the SGI gear doesn't. It says it conforms to SMPTE-292M (VBOB manual). That spec goes up to 1080i, VBOB claims 1080p ?? While there are two sets of HD-SDI BNC's for each link, I don't think it's dual link HD-SDI (SMPTE 372M). Could it be 1080p@24fps ?? Confusing

I currently don't own any 1080p or 1920x1200 capable displays, and want to resolve this before choosing one. I can do SDI <-> HDMI format conversion but resolution and refresh rate issues remain. PC display capable of PAL/NTSC or 1080i resolution? 24fps isn't uncommon in TVs, but monitors? The Dell 2407WFP I have at work can do it via analog component in, but monitors supporting that on digital inputs? No clue ...

Or maybe I should go for a small HDMI equipped TV screen rather than a computer monitor.

I'm tempted to simply connect the Onyx2 and Tezro VBOBs and make one capture the other (as a proof of concept). Or source a simple SDI-composite converter and attach to the mini-DV cam.

And I haven't even started with the software side of things. Well, I still have the original Tezro hard disk with a Smoke install. Smoke is supposed to do HD-I/O.

Any hints to get me started? Any idea if ShotMaker works with DM2 and/or TVO? Documentation says it was developed for the older XIO parallel HD I/O board (XT-HD?).

I can spend some money on it but cannot justify the megabucks pro equipment for my hobbyist projects.

The IBM PC Jr. is turning 30 too

Nice Challenge, with the SATA mode and the 100Mbit networking. I used FDDI for fast(er) networking in mine.

diegel wrote: This is of course a modified processor board. I was lucky to find a 200mhz R5000 CPU to replace the original 180mhz version. Since the CPU don't become hot, I run it without heat sink. The maximum temperature I was able to measure at the CPU cap was 42 degree Celsius.

I wouldn't worry about the CPU, after all it was designed to run at that speed (although probably design to work with a heatsink...). But I guess you replaced the oscillator and now you're overclocking the L2 cache? Or did you change a divider somewhere?

Did the CPU come from an O2 or something?

hamei wrote: But probably enough for two 1920x1200 @ 50hz displays

56Hz, IIRC. So close, but no cigar. Most panels wont sync to that.

hamei wrote: Is that the four-RM pipe or the two-RM pipe ? Or does the number of raster managers matter since the limiting factor is the display mangler ?

With more RMs you can have more features (pixel depth etc) and resolution too I think. But the limitation (as Recondas said) is the bandwidth of the pixel bus between the RMs and the DG. A deskside (2 RMs) can push enough pixels to max it out.

FWIW: A couple of years ago I had the opportunity to buy an Onyx2 rack. At the time I had an O2K deskside and an Onyx deskside and thought the rack might save some space (they're stacked after all). Turns out the footprint of the Onyx2 rack equals two desksides. This wasn't an improvement and I passed on the rack.

Me, I prefer the desksides. They allow you to 'taste' the architecture as good as the racks while taking up considerably less space. Which means I can have more

When you press the power button of one O300 module you expect the other module to power up as well? I don't think it works like that.

I have to press the power buttons on both of my O350 modules. Alternatively, you can use an L2 and issue a '* pwr up'.

The problem is that IRIX supports two thread models: traditional (sproc) threads and POSIX threads. You cannot mix thread models in a single IRIX process.

Several libraries in IRIX use sproc threads internally, probably because they date back to IRIX 5.3 or early 6.x days. For a developer this means that if you decide to use this library, and your code is threaded, it must use sproc threads.

Needless to say, these days everything assumes pthreads.

I guess you should run against /usr/lib32/*.so and grep for sproc ?

On my Origin 350 (IRIX 6.5.30) these libraries show up:

Several of these belong to the WorkShop debugger. Before you ask: I don't know why libmp_pthreads.so and libpthread.so match

But man(5) pthreads is clear:

I have an Indigo2 I don't need anymore. R10K, 175MHz, Solid Impact, unknown memory. Hinges of front 'door' broken, otherwise OK condition. No power supply .

Location The Netherlands. Beware: these things are quite heavy.