Slashdot Mirror

Well there is always Scyld, which is a more integrated and cheaper solution to SSI systems.

If you want an "easy to use" linux cluster try Scyld, it's pretty much plug and go integrated linux cluster solution out of the box. Then again, that's what you would expect from a distribution tailored explicitly for clustering and HPC solutions.

Jedidiah.

Interesting, I did a comparison. Now, I happen to own a Via Rhine based NIC so I did a vague search on the string "Linux via-rhine" (without the quotes) on Google, Teoma and Vivisimo. The string doesn't imply anything except that I want my results to show something about Linux related to via-rhine (or vice versa).

In the Teoma Results the first hit of 51,600 was a forum post where someone asked "Trying to install LMD 10 using a Via motherboard with onboard Rhine NIC configuration asks for additional parameters - anyone know these please?" It was about someone having problems with an onboard Rhine chip on Mandrake, very short, not much detail and not particularly interesting (nor would it have been helpful even if I was having problems with it).

In the Vivisimo Results the first hit of 51,600 was to a mailing-list post where the topic was "VIA Rhine problem in 2.4". Someone was having an obscure problem with a D-Link dfe-530tx, probably not what I'm looking for. Ironically there was a link in that post to what turned out to be the first hit on Google, and the mailing list post was actually an answer from a company employee at Scyld Software, which brings us to Google..

In the Google Results the first hit of 99,400 was to "Linux Drivers for PCI Ethernet Chips". The link was to a page at Scyld Software, a Linux company. It had information about several Linux kernel drivers (including Via Rhine/II) along with usage instructions, module settings, support options and diagnostic programs - not to mention a direct link to the driver source code. What I could learn from this hit was a lot, including the fact that I can use the via-rhine driver to both Rhine as well as Rhine II chips.

What I found most interesting about all this was not the results (they speak for them self) but rather the number of hits. Theoma and Vivisimo had the exact same number of hits which leads me to believe that they share the same indexes but filter the results differently (Indeed, the second hit on Vivisimo was the same as the first one on Teoma). I admit, Vivisimo has a really cool interface, especially the "Clustered Results"-thing, but the quality on the hits arent nearly as good as those of Google so none of them are Google replacements, yet. Well, that's my conclusion based on this shallow test anyway.

Oh, and thanks for the links btw - they're going into my collection.

Meanwhile, Apple's been in the news with its clusters and is catering to the distributed computing with software like Xgrid and Xsan

Which could be a "late step into the game" from Apple. There are plenty of companies out tere that have been offering complete clustering solutions for Linux for a long time now. Check out Scyld for example - what they're offering, as far as cluster computing software goes, is far in advance of what Apple is currently throwing out there.

It is nice to see new players stepping up, as competition is only a good thing, but you're speaking as if Apple was what prompted Microsoft into this move. While Apple does have one big name cluster project (Virginia Tech) and it has been very successful, it's still just one project, compared to the innumerable Linux clusters out there - and honestly, have a look at Scyld, they're a good example of the state of Linux clustering: plug it in and go. The only thing easier than that for HPC is, say, Cray or SGI, who offer serious complete solution (hardware and all) systems in a plug and go fashion. Compared to them Apple supercomputing is a hodge-podge do it yourself affair.

Jedidiah.

If you're serious about building a Linux cluster and want to pay money for a preconfigured system and associated support for it, you don't go to RedHat, you go to someone who specialises in that, like Scyld. Take a look at what they offer for their OS distribution - the whole thing is designed, ground up to work on clusters and make adminustration thereof as easy as possible.

One of the benfits of Linux is that it is flexible, and can be reshaped and repackaged accrding to differing needs (in some ways that's what different distributions are all about). If you want a cluster solution, go talk to people who build cluster distributions (Scyld is far from the only one).

Jedidiah.

That would be classic.

I dunno; the only reference I can find comparing them is this, where the FreeBSD driver is said to be using a faster technique. What's your system load when network's saturated?

My machines all have NetGear FA311's, which seem to hasis0: Applying short cable fix (reg=5)ve a few of their own problems, but at least they don't have pages of pained commentary in their drivers on how awful the developers think the hardware is ;)

I'm pretty sure wake-on-lan is possible

It is - see, for example, a knowledgebase article on it - but that's "wake on magic packet" (or Magic Packet(TM)) wake-on-LAN, not the more general packet matching wakeup that some network interfaces support.

I.e., the machine won't automatically wake up when you try to ssh into it; you need to send it a Magic Packet(TM) to wake it up. A packet-matching wakeup might be able to match incoming unicast packets to the machine, broadcast ARP requests asking for the MAC address corresponding to the machine's IP address, and other packets that it would need to respond to, so that attempting to ssh into it would wake it up, without making it respond to various random broadcasts and multicasts for which it wouldn't have to wake up (e.g., a broadcast ARP request for somebody else's MAC address, assuming it doesn't have to reply to that for e.g. proxy ARP purposes).

However, wake-on-Magic-Packet(TM) might be sufficient for the purposes of the person to whom you responded; I think one purpose for which it was intended was to allow administrators to wake up sleeping machines in order to do various remote administrative operations - including the remote software updates that they wanted to do.

Scyld computing (of Don Becker fame) did a lot of work on gigabit drivers for linux, and open source as well. If you can dig up an free copy of their beowulf operation system, you can probably get the drivers from there. Their website isn't too helpful after the Penguin computing acquisition, though.

Try this page.

Scyld computing (of Don Becker fame) did a lot of work on gigabit drivers for linux, and open source as well. If you can dig up an free copy of their beowulf operation system, you can probably get the drivers from there. Their website isn't too helpful after the Penguin computing acquisition, though.

Try this page.

That was my immediate reaction too. A bit of digging and I found out that the next firmware rev of the SqueezeBox will incorporate Wireless Wake-on-lan, so your mp3 server can be running on very low power when not in use, and your slimserver (or other wireless client) can kick it into life. Nifty.

You could write a cgi-program that calls Etherwake. Or you could be even more savvy and write a cgi or php program that just sends out the wake-on-lan "magic packet" by itself. It's just an ethernet packet. Bear in mind thought that the webserver is going to need to be on the same network as the machine that you are trying to wake up or it won't work.

I did parallel code development on Sun SMP boxes. Starting up jobs, seeing what was going on, killing zombies, debugging was all easier on one system than through different boxes you'd have to ssh over to see.

Even though I was using MPI and getting ready for a distributed memory architecture for the really big runs, the development was easier on the SMP box that showed a single system image.

I haven't used things like OpenMOSIX, and Don Becker, early pioneer of Linux ethernet drivers (not many other folks can claim a complete decade of experience with Linux networking), founded a company called Scyld that sells Linux clusters with single system image.

Sometimes it's convenient to see the whole box as if it were one, even though efficient programming dictates that you become aware of the different costs of data access (network, main memory, cache, disk). Practically speaking, developing and running parallel jobs is a higher user productivity proposition on a single system image.

That link again (clickable and working):

clickable link : http://www.scyld.com/products/beowulf/software/mon te.html

( for lazy surfers)

Two-kernel-monte (info here) will let you do that.

The problem is that it's unlikely you'd be able to do very much when this was going on anyways. Rebooting is probably not all that much worse a solution.

Still a nifty trick, though.

-Erwos

What about that? Will we be finaly able to switch kernels without a reboot?

I did that back in the 2.2 days with monte. Later with 2.4 kernels I did a few changes, added a feature I was missing, fixed a bug and such stuff. In case you want to see it. But it was never completely stable and lacked SMP support.

kexec might be a better alternative. AFAIK it is being maintained and might even have made it into the 2.5 kernel.

Here's a bit of info on monte. Looking for "monte-mips" will get you nowhere, unfortunately. And of course, you can get the source from the alt.org link posted above.

IIIA. General
The Speedo3 is very similar to other Intel network chips, that is to say
"apparently designed on a different planet".

let's see it was in the evening, so it might have been on his own time. i worked on it in the evening because i also had a job. he worked for nasa and then started his own company. perhaps it was tax dollars helping me...

I wanted to start using crypto-enabled Ethernet, only to find that Donald Becker has not made drivers for these and that he asks people to directly contact 3Com or Intel for their non-GPL drivers instead. What's preventing Don from writing his own GPL drivers for those cards? Is there some US crypto export restriction law that directly forbids it? The same condition appears to affect several Gigabit cards too: please contact the manufacturer for their non-GPL driver. What's the deal?

According to this page, the copyright for the rtl8139too driver, a substantial portion of which is your code, was claimed by Jeff Garzik illegally and fraudulently. Is this true? If so, have you done anything in an attempt to get the situation resolved? Do you think that other free software authors should be paranoid about protecting their copyrights?

RealTek!

You changed to scyld where the main objective is to earn money from the application of high-performance computing. You still make all those drivers available and update them (many thanks for that) but the company also has to make money, you need to pay your meals and your home.

What made you change, and how do you feel about that change now it's been a few years.

You changed to scyld where the main objective is to earn money from the application of high-performance computing. You still make all those drivers available and update them (many thanks for that) but the company also has to make money, you need to pay your meals and your home.

What made you change, and how do you feel about that change now it's been a few years.

While trying to get a RealTek 8139 card to work, I happened upon this. Has anything happened to resolve the situation since this was posted?

However, there can't have been much demand for it, because development ceased back in 2000.

Check out scyld.com for their beowulf distribution. It does exactly what you need (though there is a need for a dedicated head node system). You can run all of the slave nodes entirely diskless, and control booting into the beowulf stuff via a floppy, cdrom, or the hard disk.

You can pick up the CDs at cheapbytes (I think), so it's only a few $$$ for a basic install. Support or buying the professional edition will cost you bigger $$$.

The advantage of this is that all of your files and state and management is all done on the head node. The slave nodes boot up and pick up their configurations from the head node and go. Scyld beowulf is also significantly easier to install/maintain than rolling your own.

New enhancements on the latest Professional Release include full Alpha support including simple x86-like installation, full Myrinet and Gigabit support, a robust batch queuing system, web based administration and job monitoring, advanced hardware health and status monitoring, Parallel Virtual File System (PVFS) support, PVM support, PXE support and more.

from http://scyld.com/page/products/

New enhancements on the latest Professional Release include full Alpha support including simple x86-like installation, full Myrinet and Gigabit support, a robust batch queuing system, web based administration and job monitoring, advanced hardware health and status monitoring, Parallel Virtual File System (PVFS) support, PVM support, PXE support and more.

from http://scyld.com/page/products/

So exactly how do you change the kernel without rebooting?

Try the Two-Kernel Monte.

There will be some other costs, like there will need to be a hard disk somewhere for these things to boot from, but no other major costs.

Download and compile eepro100-diag.c from the Scyld website then follow Donald Becker's instructions here.

Download and compile eepro100-diag.c from the Scyld website then follow Donald Becker's instructions here.

This was a usefule article. I now know i have to write:

"image a Scyld Beowulf cluster of these." Whenever they release a new device that can run linux.

No, I want to use the 2.2.x kernels, 2.2.19 exactly. The problem is that Debian's installer for Potato is a 2.2.17pre7 (or something like that....), and I didn't have good luck getting the onboard controller to work durring the install.

Also, at the time I was doing this perticular install, 2.2.17 was the very latest kernel. The only way the onboard controller would work was with the latest netdrivers package from Scyld

This has all been rectified now with newer kernels, but if your installer uses an older kernel you're looking at updating the installer or finding a different NIC for the install. (Unless you're using a CD or Modem for install, I do network installs, so I need a working NIC during install.)

Other than that, I haven't had too much trouble with them other than the Windows 2000 bug and having them not respond to WOL (Wake On Lan) requests, even though WOL was enabled.

Scyld is a software company so they don't sell directly. To get a turnkey system they have partnered with hardware vendors who will sell a pre-integrated cluster along with hardware support and Scyld support for the integrated Beowulf Professional Edition software. Visit their site at:

Scyld Computing Corporation

Or go directly to their partners:

Compaq

PSSC Labs

Penguin Computing

eLinux

Aspen Systems

Custom Fit, Inc.

Atipa

Dan

Check out Scyld [scyld.com]. If I'm not mistaken Donald Becker (one of the founders of Beowulf) is the head of the company ... or at least has something to do with it.

My computer already supports being turned on from three sources:

1. The switch on the front
2. Magic Packet Wake-On-LAN (obligatory Linux-friendly link)
3. Time based, as set in the BIOS

At 8.45a this morning my computer powered itself on because I set the BIOS up to turn it on at that time.

With regards to the second option, I can SSH into the server on the LAN beside my computer and send it a magic packet it wake it up.

As for turning it off ... I just have to click the right buttons and it will shut itself down.

I moderate this buzzword compliant piece of electronics -1 redundant.

I ask this because one of the USB ethernet devices he supports (the CATC device) is not, as far as I can tell, included in the 2.4 kernel (although the pegasus USB-ethernet device is, but I'm not sure if that's his or not). Does anyone know why there is this split?

See above for my obligatory Beowulf reference...

I ask this because one of the USB ethernet devices he supports (the CATC device) is not, as far as I can tell, included in the 2.4 kernel (although the pegasus USB-ethernet device is, but I'm not sure if that's his or not). Does anyone know why there is this split?

See above for my obligatory Beowulf reference...

What would be really cool is a way to upgrade your kernel without rebooting, like QNX...

If a compute node takes two minutes to boot, you can't risk it being powered off when you need it. If it only takes a few seconds for a compute node to rejoin the cluster you can freely power down idle nodes. If you look at the average utilization for medium sized clusters, you'll see that this could save a lot of power.

Scyld Beowulf

What would be really cool is a way to upgrade your kernel without rebooting, like QNX...

http://www.scyld.com/products/beowulf/software/mon te.html

It's called Two Kernel Monte. It's a module that loads a new kernel into memory, does a little do-si-do dance to get it where it needs to be in the mode it needs to be in, then simply jumps to it. (Yes, you need to do

umount -a;mount -o ro,remount /

before doing this, since it doesn't do any of that...) No, it's not an in situ change, since all your processes die, but for systems with multiple SCSI or RAID cards, it can mean the difference in a few seconds for reboot and a few long minutes....

On a side note, make sure and grab my link; the one on Google is wrong (until they fix it), and it took me a while to find the right thing...

--
Never assume TFM is right.

Much of it has to do with the GPL's rather severe restrictions on which _non GPL'ed_ programs can talk to GPL'ed programs. Many of the arguments seem to stem from the fact the FSF calls GPL'ed software 'free' when its is the most restrictive license short of fully proprietory 'non-free' software. Its interesting to note that some so called 'non-free' software, while not being 'free as in beer', offers much more freedom. For instance, you can buy commercial libraries that offer much more liberal linking/runtime policies than any GPL'ed so called 'free' software would.

For a long time you could (may still be able to) write unencumbered device drivers for Microsoft operating systems without releasing source. This is _hard to do_ in linux, especially in the case of loadable modules, as the GPL folks have taken the damaging view that loadable modules are merely components of a work, this is a stance that has always mystified me, of course, it may be just that linux is less modular than I thought, I'm not a device driver writer :-)

So I think its just about the FSF's 1984 style 'doublespeak' that stirs this argument up everytime, the GPL is only free as in beer, not that that isn't a good thing mind you :-)

2 cents.

Every time there's a major new kernel release, there's whining on Slashdot about killing uptime, and how much it sucks to have to reboot to get a new kernel. Well, whine no more.

Erik Hendriks at scyld.com brings us Two Kernel Monte, a 'kernel module which allows Linux to load another kernel image into RAM and restart the machine from that kernel.' The only major limitation appears to be that it will not work with SMP machines. Apparently Erik got the idea from the Linux Bios Project.

This is not an in-place kernel upgrade tool. (That's very difficult, especially without stable data structures and APIs. It's possible, but don't hold your breath.) You are avoiding the BIOS initialization sequence (thus saving some time), but you're still rebooting. The new kernel doesn't know that it was loaded by a Linux system rather than LILO. The old kernel is clobbered (including all its data structures, device drivers, running processes, etc.) and is no longer running. This still kills your uptime and requires a reboot to get a new kernel. (Feel free to keep whining.)

This project is the Linux equivalent to the DOS program LOADLIN, which loads a Linux kernel (like LILO does) and executes it. The kernel then takes over the machine and starts booting. "Two Kernel Monte" does exactly the same thing, complicated by the need to disable paging, relocate the newly-loaded kernel into a contiguous memory block (in protected mode), return to real mode, and ask the video BIOS to reset the video mode to 80x25 text, all before executing the new kernel. As with LOADLIN, the new kernel will take over the machine and begin booting. Whether started by LILO, LOADLIN or Two Kernel Monte, the end result is much the same -- the new kernel ends up with complete control over the machine (without knowing or caring how it got there), and the old OS (if any) is moot.

This may seem useless at first glance, but it's a perfect tool to test a new kernel before installing it. Also, the full power of a running Linux system would be available to obtain (or create!) the kernel image to be booted, which could enable some interesting tricks that would be difficult (or impossible) to implement in LILO. It should also be fairly easy to boot a kernel from a complex filesystem. (e.g. NFS, Coda, AFS, etc.) It might also allow for more complex boot-time decisions, such as dynamically selecting the best kernel based on complex criteria. In principle, it should be as easy to execute alternative operating systems in place of Linux, but I don't know if this works. Any form of compression or encryption (no matter how complex) could potentially be used, even if virtual memory is required to decrypt or uncompress the kernel because of limited RAM! (Try that in LILO!)

In short, this is a cool project, and potentially more useful than it may first appear.

just after the slashback with Two Kernel Monte comes the announcement for a new kernel threatening to obliterate uptimes around the world. :-)

"It's also really efficiently written, and almost completely modular (as opposed to Linux' monolithic-plus-kitchen-sink approach)."

I don't suppose you've ever used Linux, since lsmod, rmmod, insmod, depmod, modprobe are all parts of the wonderful world of mapping things dynamically into kernele space. With it, I can have hot plug PCMCIA, USB, and other devices without having to have this monolitihic kernel you decry. There's even a Microkernel Linux which adds the features that the HURD and QNX have, that Linux doesn't.

QNX is targetted at embedded devices. That tight focus lets it get away with not having support for 64gb of ram, scaling to 8-way SMP, NUMA, swap space, implementation of device drivers for PC and non-PC hardware of all kinds, and other requirements of being a kernel that can be embedded in devices lacking a memory controller, to massive SMP systems, to computing clusters.

"QNX is much more suitable for PDAs and otehr small systems than is Linux."
Ever heard the phrase "jack of all trades, master of none?" Linux is the swiss-army knife kernel. It may not be as good for skinning as a proper hunting knife, but it also has a saw, a can opener, a magnifying glass, and many other useful tools that you might need some day. Think about it.

--

Scyld (you know, beowulf? A top news item today?) have a package on their site called Two Kernel Monte (by Erik Hendriks which says the following:

Two Kernel Monte is a kernel module which allows Linux to load another kernel image into RAM and restart the machine from that kernel. The loader supports initial ram disks and passing arbitrary kernel command line parameters to the new kernel. This version supports Linux 2.2.x and Linux 2.3.x.

Not quite the same thing, but related and interesting.

Scyld (you know, beowulf? A top news item today?) have a package on their site called Two Kernel Monte (by Erik Hendriks which says the following:

Two Kernel Monte is a kernel module which allows Linux to load another kernel image into RAM and restart the machine from that kernel. The loader supports initial ram disks and passing arbitrary kernel command line parameters to the new kernel. This version supports Linux 2.2.x and Linux 2.3.x.

Not quite the same thing, but related and interesting.