Slashdot Mirror
An Interesting Boot Log On Alpha
Here is an interesting Boot log on an Alpha. What is so interesting about this boot log? Nothing special, just that this Alpha's got 31 Processors, 256GB RAM -- looks VERY impressive. I wouldn't mind having one of those beasts at work *drooling all over*. Oh, and it compiles the kernel very fast :)
But its definately not the same system as the new 12,000 cpu 30 TeraOp ASCI EV68-based dream machine.
Now what would be _cool_ is if during checkout someone were to try out 2.4.0-test103 and find out that it actually outperforms Tru64 for certain classes of problems.
The reference to 2.4.0-test103 is based on the rate of test kernels, and the projected delivery time of this new machine. I hope that 2.4.0 final is available much sooner than that.
About 3 stories ago. I wish Compaq would commit to keeping the line running, at least for a few years. It'd make my 64 bit system purchasing decision a lot easier...
I'm trying to teach myself to set people on fire with my mind... Is it hot in here?
Yes, Linux is cool and performs fairly well. But Tru64 UNIX is optimised for Alpha hardware and far more scaleable. Yes, Linux is more scaleable than ever before, but still not as scaleable as most UNIX systems. It seems like a waste to spend all this money on excellent hardware and then stick Linux on it, degrading the performance.
ACHTUNG! ALLES LOOKENSPEEPERS!
Das computermachine ist nicht fuer gefingerpoken und mittengrabben.
Ist easy schnappen der springenwerk, blowenfusen und poppencorken mit spitzensparken.
Ist nicht fuer gewerken bei das dumpkopfen.
Das rubbernecken sichtseeren keepen das cotten-pickenen hans in das pockets muss; relaxen und watchen das blinkenlichten.
Oh, wait, front panels went out in the 70's. nevermind.
"Linux pthreads" is really just a library implemented on top of the Linux kernel clone() system call. The clone API scales just fine. The pthreads library on the other hand has "issues."
Well... the S/390 port runs on a VM, so Linux doesn't know the kind of power that is actually down there..
Nope, the S/390 port runs either in a VM or natively. You can boot up a 390 with Linux as your only OS, and then it definitely knows about the whole machine.
It's just that far more people are likely to have access to a VM running on a 390 than there are that have a whole 390 to play with.
No, no, no. It ain't ME babe,
It ain't ME you're looking for.
-- Alastair
As taken from yet another Kernel Traffic Post:
Linux is king-of-the-hill in SpecWeb99 tests on One, Two, Four, and *8-way* systems. :-)
Insanity is the last line of defence for the master diplomat. But you have to lay the groundwork early.
Starting system logger: Warning: /boot/System.map has an incorrect kernel version.
/proc/sys/net/ipv4/ip_always_defrag is missing -- cannot control IP defragmentation
Now which processor is doing that? Is it the RAM strips? Oh wait.... it's the modules...
Ok, I don't mean to troll, but if I want to show-off my machine, I'd like to fix those little warnings first before sending the dmesg.
---
dd if=/dev/random of=~/.ssh/authorized_keys bs=1 count=1024
Anyone that doesn't believe that Linux scales, is mistaken. Anyone that professes the same is either ignorant of what's been going down for the past two years or is FUDing.
While I'll admit that the S/390 port is some proof of scalability- every little drop helps dispell myths.
I am not merely a "consumer" or a "taxpayer". I am a Citizen of the State of Texas
Holy cow. My first kernel took about 8 hours as I recall, on a 386-16 with 4MB. I was struggling with signal 11s the whole time, too -- the motherboard was fine in DOS, but flaky in 32-bit mode. I'd have to restart the whole process every half-hour or so.
:)
.90 series. I think Linux has grown just a wee bit since then. I bet it would take 15+ hours for a modern kernel on that old machine. No way to know, though -- I gave that machine away long ago.
Damn, what a difference. 8 hours (or 10 in your case) versus 20 seconds. Shit, you wouldn't even have time to go get a cup of coffee anymnre. Used to be you had time for a beach trip.
Down with enhanced productivity!
Oh, just occurred to me -- the kernel I was compiling was in the
It can play an hour's worth of MP3s in 78 seconds.
(and execute an infinite loop in less than 3.5 minutes)
Duct tape + WD40 => DevOps
ttyS00 at 0x03f8 (irq = 4) is a 16550A
All I need is another 255.8725 GB, another 29 processors, another 7 LAN cards, and I'm right there!
It depends upon the task being performed of course. Some tasks can never be parallelized at all, regardless of how many CPUs you have and what operating system you are running.
Sure, but assume a task that is 100% parallelized for the purpose of the question. What is the OS overhead as you increase processors?
--
1990: You could easily read everything in comp.*. You bitched about all the weenies clogging up the alt hierarchy.
1995: You could easily read everything in comp.lang.perl. You bitched about all the weenies clogging up the comp.* hierarchy.
1998: You could easily read everything on slashdot. You bitched about all the weenies clogging up Usenet.
July 2000: You could easily read everything on kuro5hin. You bitched about all the weenies clogging up slashdot.
September 2000: Bloody weenies clog up kuro5hin. End of universe as we know it. Film at eleven.
aboot: switching to OSF/1 PALcode version 1.75
aboot: booting from device 'SCSI 3 6 0 1 100 0 0'
This joke was aboot as funny as Canada's military presence, eh?
I could not justify my existence if I were a turkey farmer. Would I terminate myself? Undoubtably, yes.
How much software actually scales well to that many processors? I suspect quake would top out long before 31.
Thank you for not thinking.
SuperID
Actually, to be more accurate, this computer was purchased by the DOE for the Sandia National Laboratories to simulate nuclear explosions. What the truth is is that if you read the entire article the computer will actually be running a modified version of RedHat Linux and their proprietary, but soon to be open source software, called CPlant. The article has all the links and it an interesting read. Check it out. More great Linux headlines that should have probably been their own post on Slashdot.
At one of my jobs, I was in the computer room of a Vast Fianancial Company That Shall Remain Nameless. They had an IBM RS/6000 SP-2 supercomputer that filled most of one wall, 32 POWER3 processors and untold terrabytes of storage.
I was left alone to work my mojo on a much smaller Sun server, but once I was good and certain i was alone, and that there were no cameras monitoring me, I wandered over to the supercomputer. I looked at the gray and blue tower that held the processors and RAM. It was worth $20, easy.
I touched it, caressed the cool metal of the mesh grid over the airvents with my fingertips, feeling the warm air and the low buzz. I'd pay $100 to do that again.
Then I licked it.
Priceless.
SoupIsGood Food
This is a hardware, not an OS issue. And sure enough, Alpha (like Pentia) has an atomic test-and-set instruction! Good thing, too -- you need this sort of instruction to implement mutexes and other locks both in kernel and user space. There is no efficient software workaround for the lack of such an instruction.
Um, dude, Microsoft doesn't run on alpha any more.
Oh yeah, just another way that Microsoft products suck.
Fire and Meat. Yummy.
Powerful, yes, but not as powerful as THIS
--
Chief Frog Inspector
A feeling of having made the same mistake before: Deja Foobar
I have talked previously with them about their stories posted on slashdot.
They said that their machines could handle it easily...
Hetz (Heunique)
I would find it unremarkable if there was some Intel chip that had an outrageously higher BogoMIPS rating that would disappear from the "running" as soon as you tossed POVRay onto it...
If you're not part of the solution, you're part of the precipitate.
Is this a breakthrough technology? No. Is this an earth-shaking legal or political development? No. Is it something that geeks the world over will have wet dreams about tonight? You bet your ass.
All the posts about "I need a towel" and "they should sell tickets just to touch it" are gonna look funny when this is just another slag heap of unusable parts.
"Why, when I was yer age, miboy, we had to put up with using a computer. That's a complicated physically connected brick of processing components. We thought a mere 2^5 processors was worth drooling over. Yes, miboy, I know your cochlear implant has more than that. You're missing the point. This thing was tremendous! It took up a whole rack: four times the size of a grown man! And all of its memory circuits were in the same cabinet, requiring massive cooling apparatus, unlike the distributed memory crystals that people embed in their jewelry."
Anyone still drooling over 2^5 address space on ferrite core memory? Anyone still drooling over 2^5 address lines? Or data lines?
[
If the instruction code were compatible with the x86 set, how fast do you guys think this would run Windows 2000? 2 second boot time?
- I don't care if they globalize against free speech. All my best free thoughts are done in my head.
Hey coward!
Read some news! IBM, Compaq, Dell and the other guys are WORKING on making 16 processors and 32 processors based server - it's still vaporware!
Also, do you mind showing me a X86 machine which supports 256GB RAM? huh?
Thought so!
Hetz (Heunique)
People seem to be complaining that "this isn't news".. no, it's not the kind of thing that you'd see on your local TV news, but it's "News for Nerds". Which is, as I'm sure you're aware, what this is all about. Why should I care that you don't like this story? I didn't like the last one, do you care?
-
Meep meep
> Can anyone comment on the SMP performance linearity of the current Linux kernel on more than 4 CPUs?
It depends upon the task being performed of course. Some tasks can never be parallelized at all, regardless of how many CPUs you have and what operating system you are running.
> Did they every sort out the issues that prevented kernel socket (or was it I/O?) APIs
> being called concurrently by processes on multiple CPUs?
Yes, Linux TCP/IP is fully threaded and will run concurrently on all CPUs, assuming that there is work to be done on all of them. (a single socket will not run on 100 CPUs at the same time, for instance)
This means nothing. Being scalable means that performance with 30 processors is about the double of performance with 15 processors, and so forth. If this 31 processors machine, running Linux, can deliver more juice than a machine with 8 processors, I'd be surprised.
(8-DCS)
Processors 1 to 15 would work on distributed.net...
:)
Numbers 16 to 20 would do Seti.
Numbers 21-29 would run Quake3, Civ:CTP, and XWS.
Numbers 30 and 31 would run the realtime disk encryption/decryption series
I used to be someone else. Now I'm someone better.
Real life is underrated.
Actually I thought it was kind of cool. Big iron is interesting. Running Linux on big iron is very interesting. :-)
I wonder how long kernel compiles actually take?
Alpha chips were a dead technology? Keep saying that, it'll drive the price down! OK, an Ask /. question here: What MPI tools are available for Linux, like LAM? IOW, how does this clever guy do useful things with his Moonbase Alpha?
I am quite civilized, and I should be brought a beer immediately. -- Bruce Sterling
that *still* depends on the workload. if it's CPU-bound, it just works. if it's I/O bound, it depends on global system recourses (bus speed and bandwith), and on how well smp-threaded that particular subsystem is on the OS. one of the big improvements of the linux 2.4 kernel is precisely a major overhaul of the tcp/ip stack to smp-thread it completely. OTOH, heavy filesystem work (lots of rename(), unlink(), open(O_CREAT), etc) is not likely to scale super-well, because of filesystem locking constraints, and that's under all OSes too, which is why you generally avoid writing your big programs in such a way that they have to be messing around with lots of little files all the time (e.g that's why you use a database backend, or some kind of db format, rather than flat files, if you want a news server that scales). beyond that, in the real high end (32 procs and more), the SMP model itself starts showing its limits, which is why NUMA (non-uniform memory access, i.e. not all memory is equally fast from each processor) was invented. Linux 2.4 has some preliminary support for NUMA, but it's still in the beginning stages.
that's one side of the story. the other side is that clone() doesn't support the exact POSIX threads API, and the linuxthreads library has to bend over backward (and be slow, in the process) to add this support. the situation is slowly fixing itself, but there are some human communication problems in the middle, and it takes a lot of time for a new version of linuxthreads to make it all the way to the distributions.
If you scroll down and pause at certain intervals you can pretend its your computer booting
So is this the Jugs magazine centerfold for kernel hackers? "Look at the processors on THAT!"
Computing speed is relative. You can buy a faster computer, or make your mind go slower. Bring out the beer!
The TCP/IP code is fully SMP threaded. It runs on multiple processors in parallel. The same is true for the VFS layer. Linux already holds the world-record SPECweb99 score on a 4-way box, which would not have been possible without good SMP scalability.
"Feh" is right. Who needs an alpha when I can get 0.0033 Bogomips on my c64. Read it and weep!
Imagine a beowulf cluster of C64s! Think of the power!
Here is what might have just happened : /. and decided to focus our attentions by confusing us with that headline while he's kidnaping CmdrTaco and Hemos.
1) This is one heck of a boring day with nothing happening at all.
2) HeUnique just received the rights to post and wanted to post something, anything quickly before those rights faded away.
3) Someone evil took controle of
4) I was really bored at work and decided to reply to that headline even if it didn't hold much interests.
"When I was a little kid my mother told me not to stare into the sun...
"If liberty means anything at all, it means the right to tell people what they do not want to hear"
Quoted from article:
Oh, and it compiles the kernel VERY fastGood. Maybe it'll be fast enough to start up Windows 2000 in a reasonable time frame.
<grin> Think of how much *ass* you could kick running SETI@home on that!
Fire and Meat. Yummy.
imo, what it can do well is less important than what it can't. To paraphrase Limp Bizkit, "Let's break some sh**."
I don't want knowledge. I want certainty. - Law, David Bowie
its porn for geeks! if you will excuse me... i think i need a towel now.
At a guess, not very long at all. On a 24 CPU Sun Starfire machine, a kernel compile takes a shade over 20 seconds. See http://linuxcare.com.au/anton/e10000/ for details. I'd guess this Alpha will be comparable to that, if not faster. As an interesting datapoint, my first kernel compile took over 10 hours (that was 0.99pl8+ on a 386).
"The invisible and the non-existent look very much alike." -- Delos B. McKown
I can't resist saying this.... it'd be cool to have a beowolf cluster of these babies!!
Can anyone comment on the SMP performance linearity of the current Linux kernel on more than 4 CPUs? Did they every sort out the issues that prevented kernel socket (or was it I/O?) APIs being called concurrently by processes on multiple CPUs?
Indeed, but other (more competent :-) companies are already
there, even with Intel CPUs. The Data General
AV25000, for example,
supports up to 64 PIII Xeon CPUs, and runs either DG/UX or Windows (or both). Of course, the
if you go the Windows route, you'll have to run multiple copies simultaneously, because Windows can only
scale to 4 CPUs on that machine. If you go the more sensible DG/UX route, of course,
it can use all 64 CPUs from a single system image :-)
"The invisible and the non-existent look very much alike." -- Delos B. McKown
I did a search for "Bogo" to find the BogoMIPS and found this:
Calibrating delay loop... 1488.98 BogoMIPS
Feh, I thought, either Alpha's REALLY suck or (more likely) there's a bug there. Then I took a closer look:
SMP starting up secondaries.
Calibrating delay loop... 1493.17 BogoMIPS
Calibrating delay loop... 1493.17 BogoMIPS
Calibrating delay loop... 1493.17 BogoMIPS
Calibrating delay loop... 1488.98 BogoMIPS
Calibrating delay loop... 1488.98 BogoMIPS
Calibrating delay loop... 1488.98 BogoMIPS
Calibrating delay loop... 1488.98 BogoMIPS
Calibrating delay loop... 1488.98 BogoMIPS
Calibrating delay loop... 1488.98 BogoMIPS
Calibrating delay loop... 1488.98 BogoMIPS
Calibrating delay loop... 1488.98 BogoMIPS
Calibrating delay loop... 1488.98 BogoMIPS
Calibrating delay loop... 1488.98 BogoMIPS
Calibrating delay loop... 1488.98 BogoMIPS
Calibrating delay loop... 1488.98 BogoMIPS
Calibrating delay loop... 1488.98 BogoMIPS
Calibrating delay loop... 1488.98 BogoMIPS
Calibrating delay loop... 1488.98 BogoMIPS
Calibrating delay loop... 1488.98 BogoMIPS
Calibrating delay loop... 1488.98 BogoMIPS
Calibrating delay loop... 1488.98 BogoMIPS
Calibrating delay loop... 1488.98 BogoMIPS
Calibrating delay loop... 1488.98 BogoMIPS
Calibrating delay loop... 1488.98 BogoMIPS
Calibrating delay loop... 1488.98 BogoMIPS
Calibrating delay loop... 1488.98 BogoMIPS
Calibrating delay loop... 1488.98 BogoMIPS
Calibrating delay loop... 1488.98 BogoMIPS
Calibrating delay loop... 1488.98 BogoMIPS
Calibrating delay loop... 1488.98 BogoMIPS
SMP: Total of 31 processors activated (46170.90 BogoMIPS).
Oh.
--
Non-meta-modded "Overrated" mods are killing Slashdot
(Hey Ryan! Here's your proof!)
I'm glad you almost guarantee it, because I'd almost like a different opinion. I mean, if it's a "good" salary, I damn well better be able to afford 32 CPUs in a single machine!
/. is that machines like that make geeks drool. The fact that it compiled and booted is exciting in the same way that people who live in an area get excited when the local ball teams win games. It's not really rational, it's tribal bonding. Reports like this confirm the Slashdot tribal belief that while Linux may never be Grandma's desktop OS, it will continue to make inroads and eventually dominate the real computers of the world. This makes the Linux enthusiasts very happy, especially since financial rewards are secondary to the more important free-software geek reward of enhanced reputation (so saith ESR at least), and this is like reputation enhancement by mere association. "Sure I only run Red Hat on my aging P/75, but if I had 32 CPUs and a computer the size of an elephant, I could still run Linux. Let's see you do that with Windows 2000. Nyeah!"
For the record, I don't think the linked page is a brag. I think it's notes to interested parties who are working on porting the Linux kernel to large multi-processor machines. The reason it got on
I do not have a signature
Look at the "active CPU mask" in the log. CPU 7 is offline.
Hey, welcome to Retro-Slashdot! Back before there was all this legal coverage, back before "grits" were ever mentioned, back before moderation reached the masses, back before user customization of displayed stories, even back before the coming of the Evil One (Jon Katz) . . . THIS is an example of what Slashdot regularly did.
Steven E. Ehrbar
The same way that script kiddies do :-)
You used the equivalent of a megaphone on slashdot (the +1 bonus) to tell us that you like it and think it's a good thing, in only slightly more words? Did you think this was actually insightful or worthy of everyone's attention, or did you just think "I want everyone to know what I think, and though I'm not more sophisticated about my thoughts than everyone else without a +1 bonus, I'm going to go ahead and say it, because I can"?
.sig is not generated by each user when he posts a comment, where he's free to forge it. It's appended by slashdot's servers. It's an effective deterrent to impersonation, and it must again receive its prominence within the social jurisprudencial realm of slashdot.
There is an aristocracy on slashdot; make no mistake about it. When people like drendite (userid=#3) speak, people bow down in worship, simply because of his low userid. It doesn't matter whether one actually makes a true contribution to society; what matters is the aristocratic entitlement conferred by longstanding existence (not participation).
Take the British Parliament, for example. Though Britain still hasn't come close to providing universal healthcare or proper dentistry they sorely need, they have finally seen the folly of maintaining a ruling aristocracy, and have eliminated the hereditary seats in the House of Lords. Slashdot should follow their lead.
The solution is not to take away the voice of people with low userids. Nor is the solution to eliminate the +1 bonus, because it serves a legitimate purpose and is democratically attainable by all, from the oldest poster to the neophyte with a five-figure userid. The solution is to eliminate the tagging of comments with the userid of their posters.
The userid tag does nothing to help the community, and does much to harm it, by encouraging wishywashy moderators to inflate the karma of oldtimers and penalize the newguys who express controversial opinions. (When moderators waver between slamming a post or modding it up, they usually defer to the userid in addressing its seriousness and authority. This is unacceptable.)
If you're worried about fraud and impersonation, then you already have an effective means of distinguishing between posts: the signature. The
Effective policing (moderating) can only go so far. We must correct these social ills by striking at their sources -- their causes -- not merely at their symptoms. Join with me in tearing down the illegitimate reign of the slashdot aristocracy and their petty notions of insight and imformativeness, and lift up a glorious new tomorrow, where everyone, democratically, no matter what the tld of his email address or the number of his userid, shares in the same promise of opportunity for reasoned argument and receptive audience.
Thank you for your time.
Froid
And I'm quite aware that BogoMIPS are an even more Meaningless Indicator of Processing Speed than the indicators that people try to take seriously... It is of practical value, but only in predicting the performance of timing loops...
If you're not part of the solution, you're part of the precipitate.
I work at DEC^H^H^HCompaq. We have one of these bad-boys in a lab, and I have to walk by it a lot. It's this fairly massive thing, the size of 3 or 4 fridges all in a row. They're bluish, and the heart of the beast is identifiable by the LED display that sits at about eye height, saying things like:
AlphaServer GS320
16 processors configured
And so on. I adore it. And every once in a while, when nobody's looking..
I give it a hug.
Once, I was talking to a co-worker about it.
"You know the Wildfire in the lab?" I asked.
"The what now?" he replied. So I told him about the Wildfire. Later that day, we were walking through the lab, past it.
He gave it a hug.
If anyone else wants to send the wildfire a hug, let me know.
I think we should hold a fund-raiser in which people would pay $20 to be able to see this thing in the flesh. $100 if you want to actually touch it.
:)
I'd be first in line