Slashdot Log In
Fitting A Linux Box On A PCI Card
Posted by
timothy
on Sat Nov 03, 2001 09:42 AM
from the ultradense-to-the-desktop dept.
from the ultradense-to-the-desktop dept.
An Anonymous Coward writes: "Running on Newsforge/Linux.com is a hardware review where Slashdot's Krow took a couple of OmniCluster's Slotservers and and built a cluster configuation inside of a singe host computer (and even had DB2 running on one of the card's inside of the host). Could something like this be the future of computing where for additional processing power you just kept adding additional computers inside of a host?"
This discussion has been archived.
No new comments can be posted.
Fitting A Linux Box On A PCI Card
|
Log In/Create an Account
| Top
| 137 comments
| Search Discussion
The Fine Print: The following comments are owned by whoever posted them. We are not responsible for them in any way.
Ob Beowulf comment (Score:2)
It would be cool to have completely separate processors in a box, so that as long as there is power, each card can run on its own. Then you could network them together into a beowulf cluster, and then make clusters of clusters
the AC
Too bad... (Score:1)
Seen these for a long time (Score:3, Insightful)
The SETI version (Score:3, Insightful)
http://slashdot.org/article.pl?sid=00/07/23/215
It would have been GREAT to have an improvement in CPU speed on a PCI card, as I always have at least two free in every system I own. What I wonder, though, is what instructional speed would the PCI card "CPUs" give us?
Impractical (Score:2, Troll)
Also, I would imagine that the RF interference generated by having several of these in one box would be quite signifigant. PCI slots are only an inch or so apart on most motherboards, and without any sort of RF shielding between multiple cards, I can't imagine they'd function properly. It's a good idea on paper, but in reality, I'd think a few 1U rackmount servers would do the job much better. And for $499 a piece, you could get a decent single processor rackmount server for around the same price.
Re:Impractical (Score:4, Informative)
don't see these things taking off for most uses because the PCI bus is limited to a measly 133 MB/S. Even newer 64 bit PCI slots found in some servers have insignifigant bandwidth to keep the data flowing fast enough to make full use of these things.
You've heard of Beowolf clusters, right?
Let's imagine I'm running some large routine to model some physical phenomena.. Depending on the problem, it is often possible to split the computational domain into small chunks and then pass only the elements along the interfaces between nodes.. So, how does that impact this discussion? Well, let's assume I can break up an NxM grid onto four subdomains. The communication from each node will consist of N+M elements (not NxM).. Now, let's take a look at our options. I can either purchase 4 machines with gigabit (~1000Mb/s) ethernet, Myranet (~200Mb/s) cards, or maybe I can use ip-over-firewire (~400Mb/s) to communicate between machines.. Gigabit ethernet has some latency problems that are answered by Myranet, but if we just look at the bandwidth issue, then ~1000Mb/s is roughly 125MB/s. That's slower than the 133MB/s you quoted above for a 32bit, 33MHz PCI bus.. Of course there are motherboards out there that support 64bit, 66MHz PCI cards (such as these from TotalImpact [totalimpact.com])..
You're right that the PCI bus is not as fast as the data io approaches use by IBM, Sun, SGI, etc to feed their processors. BUT, if I'm deciding between one machine sitting in the corner crunching numbers, or 4 machines sitting in the corner talking slowly to each other through an expensive gigabit ethernet switch, guess which system I'm going to look at?
Geode Specs (Score:2)
CPU Speed (Score:1)
Linux on PCI cards is the way forward. (Score:1)
Just hand them a PCI card and let them get on with it. I can't help thinking it would be better on a USB device though. Then you wouldn't even need to open the case !
Isn't that the course we've been on? (Score:4, Funny)
A computer used to take up a room.
Then, computers were large cabinets in a computer room.
Now, they are boxes in a computer cabinet in a computer room.
So we can extrapolate the next step for computers is to be cardss in computer box in a computer cabinet in a computer room.
It's a natural (obvious) progression really.
future? yes, but it's here today... (Score:1)
newest systems are almost this exactly, but instead of slow, thin pci, they use large, fast
interconnects:
http://www.sgi.com/origin/300/
Wait! What about Beowulf? (Score:1)
Here we have four or five cpus all in one machine, talking to eachother over a native PCI bus. It seems to me this would be a great way to run a Beowulf cluster In a machine.
Anyone care to comment on why he might not have done this?
If you do have storage issues... (Score:1)
These cards have been around for ages with various degrees of complexity. There used to be (don't know if they are still around) some of these cards that were designed to plug into a Mac so the card would do all the hard work if you wanted to emulate a PC.
I don't see the value for the home user. I can't see why a true home user (not the very small percenteage of hardcore enthusiasts or people that run a business from home) would need so much power that the solution is to get a box, plug a few of these babies and cluster them.
Still, its not so hard to come up with a home scenario:
1. Send your broadband connection to the basement of your house and spread it to all the rooms in the house with a $80 broadband router, cheap switches and hubs.
2. Put a box in a closet in the basement with different PCI cards to serve a specific purpose. For my own personal needs (I am a Microsoft dot whore, sorry) I would have an exchange server, one dedicated as a network file server, a sql server and a IIS server. A person of the Unix persuasion would have a card with sendmail and some kind of pop server, a file server, mysql or posgres and Apache.
With just a little bit of money the house now packs as much punch inside of that box in a basement closet than what it takes my company to do with a row of bulky servers. Add in a blackbox switch and a cheap 14-in monitor, keyboard and mouse and you are set. Of course Unix people would use some kind of secure shell and save themselves the trip to the basement, and us lazy Microsoft whores will just have to rely on Terminal Services or pcAnywhere.
In a corporate environment the space saving actually pays off (you don't pay your apartment rent or home mortgage by the square foot like most businesses do) as soon as you recover some of the space wasted by the server room. Right now I can see how I could take ours, gut it out, put a couple boxes full of these PCI cards in a good closet with the proper ventilation, and then turn the old equipment room into a telecommuter's lounge.
The home solution would rock because my wife will not bother me anymore about all those weird boxes sitting under my desk in my home office. All the clutter goes away and I just keep my tower case.
Geode? (Score:2)
Too bad a Dual Athlon-based solution (on a full length PCI card) would suck too much juice... at least from the current PCI specs... AMD needs to make a move like intel did with their Low wattage PIII, I'd love to see a 12 processor (5 pci slots plus host) renderfarm in a single box for a decent price. Not only it would be space saving, but imagine that in a plexi-glass case
Wait just a minute... (Score:1)
I don't know if I can take another disappointment like that.
Imagine a new kind of bus (Score:2, Insightful)
Essentially there would be a switch that allowed about 32 devices to be attached.
The devices could be storage devices, processors, audio/video devices, or communication devices.
Storage devices would be things like memory, hard drives, cdroms and the like.
This bus would allow multiple processors to access the same device at the same time and would allow devices to communicate directly to each other, like allowing a program to be loaded directly from a hard drive into memory, or from a video capture device directly onto a hard drive.
No motherboard, just slots that held different form factor devices with power and optical wires attached.
A networking device would allow the internal protocol to be wrapped in IP and allow the interntal network to be bridged onto ethernet. This would allow the busses on seperate computers to work like a single computer. The processors on all the machines could easily network together, memory could be shared seamlessly, harddrive storage would be shared and kept backedup in real time. Any device in any machine could communicate directly with any other device in any other machine. Security allowing.
Want 20 processors in your machine? Install them.
Want 6 memory devices with 1GB each? Add them.
Want 100 desktop devices with only a network device, display device and input/output device that use the processor and storage out of an application server? No problem.
Want a box that seemlessly runs 20 different OSes each in a virtual machine that are ran across 10 boxes in a redundant failover system? No problem, it's all done in hardware.
Want the hard drives in all the desktop machines to act like one giant raid 5 to store all the companies data on? No problem. (1000 machines with 10 GB each is 10 TB of storage)
This is the future of computing.
SUN has a similar product.. (Score:5, Interesting)
//Phizzy
hmmm.. (Score:1)
How does sharing of the disk between each machine on a card affect the performance ?
Audio Apps -- Digidesigns DSP Farms (Score:2)
But anyway, it reminds me a quite a bit of what Avid/Digidesign do for their high-end systems.
You see people who've got 6 slot PCI systems and 4 of those slots are filled with extra computing cards (sometimes more... some people get expansion chasis'). You can rely on your computers processor if you're not doing to many complex effects on a track of audio, but at some point (not too hard to reach... throw in a tube amp emulator and a reverb) you run out of CPU. So they have PCI cards which have a couple of DSP chips (Motorola 56xxx series, I think) on them, and the more of these you add, the more audio processing you can do simultaneously.
At some point, perhaps people will think: hey, why add a specialized card? Why not just more general purpose computing power?
my problem is price (Score:1)
my idea setup will be using a CF card with CF-IDE adapter as the boot drive(which eliminate the dependancy of the host OS on powerup and no actual HD required).
G4 PCI cpu (Score:1)
http://www.sonnettech.com/product/crescendo_720
The Catch: You have to write the device driver for the Motorola MPC107 PCI bridge chip.
Switched Bus, Multipurpose cards (Score:2)
For cards, it'd be great if each card had its own CPU and RAM. Ideally the cards would have a few universal connectors, each of which could accomodate an I/O module which would contain just the discrete electronics necessary to drive a specific device or medium (eg, video, audio, disk, network).
Bus-Switch modules would be interconnectable to accomodate more cards, and would have switch-like manegement features for segmentation, isolation and failover type features.
The CPU cards themselves ought to be less complicated than motherboards since there's no bus logic, just interconnect logic to the Switch-Bus and the I/O modules, and RAM.
Since each board has its own RAM and CPU it ought improve system performance because the O/S could offload much more processing to CPU boards dedicated to specific tasks. Instead of the kernal bothering with lower-level filesystem tasks and driving the hardware, a "driver" for filesystem and devices could be loaded on a CPU board dedicated to I/O.
The same could be true of user interfaces -- run the the UI on the board dedicated to video, audio and USB. The kernal could run applications or other jobs on the "processing" CPU board(s).
Networking? Offload the entire IP stack to the networking CPU board.
i dunno about these (Score:1)
ClearCube has another one (Score:1)
rackmounted PCs with video, etc. They're intended for offices: you run cables to each person's monitor/keyboard/mouse, manage all the actual hardware in one place ~~ ClearCube [clearcube.com]
Linux on PCI a year old, dudes! (Score:2, Insightful)
I looked at this and said... wait a minute, hasn't this already been sorta done [slashdot.org]? Despite not being a full featured box, Firecard [merilus.com] is a PCI-card running Linux... for the purposes of supporting a firewall (as you could have guessed from the name if you'd not read the story -- Nov 14 2001... but it's cool that they've taken it to the next level.
Radius Rocket (Score:1)
http://lowendmac.com/radius/rocket.shtml
Single card servers (Score:1)
I want some ! (Score:1)
This idea is actually the standard (Score:1)
You can get 4, 8, 16, or even 24 SBCs (Single Blade Computers) in a chassis, and link these chassis together via switches. Each chassis has a switch that links all the SBCs in the backplane together and has external ports to hook it up to the outside world.
Check this out:
http://www.picmg.org/compactpci.stm
and this:
http://www.intel.com/network/csp/products/cpci_
Haven't Sun been doing this for a while? (Score:1)
They demonstrated how an entire Windows NT cluster could be built using this technology, chucked in some Terminal Services under Windows, ran Exchange, and then did all the important stuff (mail, DNS, whatever) on the Sun box itself.
Granted, it's not Linux, and granted, he cost of a Sun box is quite high - but the PC cards are significantly cheaper over here for Sun hardware, and Sun architecture seems to be a bit more robust and scalable than PC stuff.
who cares? (Score:2)
Nobody cared then.
Why would anyone care now?
Please explain your point using no more than 100 words.
-
Sunpci card (Score:2, Informative)
I think you can pick them up pretty cheap nowadays if you like that sort of thing. I don't imagine much mileage from trying to install a.n.other unless you feel like writing the relevant drivers to get everything to talk to each other.
wait a minute (Score:2, Insightful)
as has been noted before, this would really be useful if the pci bus was extended (faster/wider). of course, making it faster/wider gives you what sgi has been doing for a while too (also mentioned above).
perhaps the most dissapointing thing is that all that power goes to waste on users playing solitare, running windows, aol, and quake, not on something that will actually need the power to perform the tasks. well, maybe quake isnt so bad...
days of future passed (Score:3, Informative)
No, not if it's existed for decades. It's what's referred to as a "mainframe". You know. An expandable number of processor boards running under an operating system that can treat them as any number of single-processor or multiprocessor machines, with the ability to reassign processes between CPUs.
The Unix world has had them for a long time, too. Modern examples include Sun's higher-end servers, which support hot-swappable and hot-pluggable processors and memory.
Doing it with x86 processors and standard x86 OSes like x86 Unixes and Windows is less common but I believe Compaq and maybe Unisys can sell you machines that can do it, too, with one or several instances of the OS running at once.
This hatdware approach is not quite the same as VMWare's server products, which do it via software and don't limit you to one OS per processor or block of processors. It in turn mimics other decades-old mainframe operating environments in approach.
Some merit to the design, but this is better. (Score:1)
However, Powerleap [powerleap.com], ubiquitous for upgrades and socket adapters, also has a card which touts some similar attributes called the Renaissance/370S [powerleap.com] based on a Socket 370 or FC-PGA chip. It cranks with Celeron, Celeron II, and P3 chips. Quite rockin'.
The main cool thing about this device is it does NOT use the motherboard slot it sits in. It just uses it as a place to mount. That's right, you can put one in an ISA slot and still run the motherboard it sits in and they won't know a thing about each other, because no pins are connected between them. The price is also a lot better (~$250 for a low end model), you can swap out the CPU and it has two DIMM slots, with a max ram per slot of 512mb (1GB combined). The specs are much better and the price is much lower. It's just marketed as an upgrade option rather than a performance enhancement to an existing machine.
I've been looking into this as a solution for my cluster, but haven't gotten up the nerve to buy them yet. From what I can find on the web, they're the best cluster card option, especially if you are handy with soldering. To really maximize the power per box, I'd probably buy a dead 486 motherboard (ISA slots all the way across the board, which this card requires), slam four Renaissance cards in it, link two power supplies in parallel, rig extra power and reset switches for each card separate from the power supply, and there's your mini-cluster. Probably 4 machines per 4U case, which notably isn't a huge space savings over 4 1U pizza boxes, but it costs less than a single 1U server would.
This is nothing new (Score:2)
Drivers only work with a stock RedHat kernel (Score:1)
The bundled kernel module only works with the stock kernel distribution in RedHat 6.2-7.1 (kernel 2.4.2 max). The kernel module sets up a virtual network device that allows the host pc to talk to the slotserver. The kernel module needs to run on both the host and the card to be able to communicate with each other localy. (You can still communicate via the 10/100 interface over the network.)
Another thing they advertize is the ability to have the card boot off of the host computer via a "virtual disk" (rather than having IDE drives hanging off of the card). I have't been able to get this working at all - and the only documentation available tells you that the feature exists.
It would kind of suck to have a PC loaded with 4 cards and 4 additional disks. I could sell a kidney and purchase some disk-on-chips I guess.
-Andy
Computer in a computer (Score:1)
You don't understand... (Score:1)
Re:Performance (Score:1)
Since I've got a router/firewall box using the same CPU I should be able to answer that one (300Mhz):
RC5: Summary: 4 packets (25.00 stats units)
1.14:10:07.62 - [48,778 keys/s]
As you can see, number crunching is rigtht out of the question. Still, easily fast enough to push packets around.