Oracle's GPL Linux Firewire Clustering

Smoking writes "It seems that Oracle just released libraries to allow low cost Linux clustering solutions using firewire... Aside from the coolness factor (imagine a beowulf cluster of DV cameras...) it's quite new for Oracle to release GPL software. They also seem to include really useful tools for NIC failover, Wizard building framework and integration of the cluster into Gnome (via a gnomevfs plugin)."

Thanks Oracle!

[zulux (112259)]

I was wondering how I was going to cluster a group of PostgreSQL servers!

Thanks!

Cheap!

[grub (11606)]

The Firewire cards needed to build a cluster can cost as little as 10% as much as the required FiberChannel hardware

Not to mention the FiberChannel switch. The Brocade [brocade.com] fiber switch we use to tie our three SGI Origins to our SAN's storage RAID was over CA$12K when we bought it.

Re:Cheap!

[ivan256 (17499)]

The Firewire cards needed to build a cluster can cost as little as 10% as much as the required FiberChannel hardware

Not to mention the FiberChannel switch. The Brocade [brocade.com] fiber switch we use to tie our three SGI Origins to our SAN's storage RAID was over CA$12K when we bought it.


Yeah, but you only get 20% of the speed. Fibre Channel is at 2048Mbps now, compared to the 400Mbps of Firewire.

Re:Cheap!

[JabberWokky (19442)]

Yes, but for some applications dropping $12k into the budget is not possible, and yet something similar at a lower cost would be ideal. Heck, this is cheaper than shared SCSI using brand name equipment. Not a bad compromise between speed and cost.

--
Evan

Re:Cheap!

[JabberWokky (19442)]

I'm not talking about running Oracle on top of it... if you have a project budget for Oracle, you can afford something better. Nor would I like to be the first to implement this in a production environment. :) That said, it's likely very stable, since it's on tested hardware and firewire drivers in general have been well tested. I'm thinking about situations where you'd normally have a couple servers sharing a big RAID via NFS over a small shared switch. This would fit into that level of sharing (and a few others as well), just better performance.

What's going on with firewire anyway - is there a bandwidth increase on the horizon? I tend to follow server hardware, and I know squat about firewire other than the three names and it supports 128 devices without having to have a central server a la USB. Is this is a solution that would be even more attractive when a higher capacity firewire rolls out in six months?

--
Evan

Re:Cheap!

[DuBois (105200)]

Not to mention the fact that FireWire is mostly self-configuring. I've not seen a self-configuring Fibre Channel anything.

My office-mate just spent a week attempting to configure a Brocade-switched Fibre Channel setup for HACMP. In his defense, it was his first attempt at such.

Everything I've ever heard about Fibre Channel reminds me of something Rube Goldberg threw together.

Re:Cheap!

[chef_raekwon (411401)]

they that can give up speed(fibre channel), to obtain ease of use (fibrewire), deserve neither ;)

Re:Cheap!

[DuBois (105200)]

they that can give up speed(fibre channel)

Most people running a small operation don't need the speed of Fibre Channel. 50MB/s is enough. But FireWire2 is rumored to be 100MB/s and up (perhaps even 200MB/s). Note the MB (megaBytes).

Re:Cheap!

[WatertonMan (550706)]

FireWire2 is extremely exciting. Since Apple will undoubtedly include it as standard in their future machines it could make some serious networking/mulitprocessing stuff possible. I hope that OSS takes advantage of this. How about a port of POOMA that works off a network of FireWire2 based computers?

(For those not familiar with it, POOMA is a math library for C++ that handles multiprocessing in a very easy way. Debug it on a single processing system and run it on a multiprocessing system) It was developed at LANL but a lot of people use it. With FireWire2 and a bunch of cheap systems you could get a lot of supercomputer performance very cheaply.

Re:Cheap!

[Twirlip of the Mists (615030)]

I have to confess that I don't actually know what HACMP requires at the switch level-- I've never used that particular HA implementation-- but setting up FC failover on SGI systems is about as simple as it gets. You just set a config file on the host telling it what the primary and failover device paths are for each LUN, and off you go. It requires no special configuration at all on the switch, so it's very close to plug-and-play.

Not every application of fibre channel has to be complicated.

This comment officially sanctioned...

[netsharc (195805)]

Imagine a Beowulf Cluster of DV Cameras..

• in a Girl's College Changing Room.
• taping Natalie Portman?

Oh yea

[mao che minh (611166)]

Now imagine a CLIC cluster of these....

Haha get it?! Because people are always like "imagine a Beowulf cluster..." so I said imagine a CLIC cluster! Haha! Genius!

I had hoped the firewire was for net

[tolldog (1571)]

I hoped that they were making strong firewire net connections and ways of channeling the systems together into some sort of hypercube formation.
That would make it appear as a true parallel processing system and giving some API to take advantage of it. I guess something like that is still possible and with firewire being fast and cheap, it is something that may be worth looking in to.

-Tim

Firewire's future

[runenfool (503)]

This is great news for anyone that is a firewire afficionado. Because millions of people will be doing firewire clustering? No. But it does show the versatility of the standard. Its a shame that Intel has such a hard on to kill it, because firewire really is a great technology.

As firewire begins to scale to higher speeds this looks like an even better method to connect not only things like computers and their peripherals - but things like your television to your PVR to your camera to your computer.

So few new firewire products

[burgburgburg (574866)]

Has anyone else noticed how few new firewire products have been introduced over the last year? How everything seems to have been routed into USB 2.0?

I'd point out how unfriendly that is to us Mac users, but somehow, I don't think they care.

Re:Firewire's future

[afidel (530433)]

Firewire will never die. The reason is simple, the next generation of midi is based around firewire as the physical transport. The origional midi has lasted decades because it was well designed and met most of the needs of those using it, but more modern setups that need to send actual sound data, not just notes were poorly supported with the old standard. Now there is a midi standard that can do everything the old one can and also carry samples!

Re:Firewire's future

[Mattsson (105422)]

Mmm... This is probably a bit off topic, but I've always wondered *why* there isn't a midi over ethernet standard.
It would be great if I could just simply plug my synth's into my network and assign ip's to them.
That way I wouldn't have to have a *separate* midi "network" and I would be able to use both my mac and my pc to make music without having to move my midi interfaces between the two. :-/

And with gigabit ethernet availible, there's no problem with the bandwith being to small...

Also, you wouldn't have to connect everything *exactly* like before when moving them.
It really is hell to get everything back together in a working fashion when you've been out and about with your synth's... :-(

Re:Out of the political loop

[runenfool (503)]

There are a couple of reasons Intel might want to kill firewire.

1) It was designed by a competitor, Apple (who made the situation worse by implementing a high fee for a time). If it was designed by say, Microsoft or Dell, I doubt they would be working so hard to marginalize it.

2) It uses no CPU resources like USB. Greater tax on CPUs = need for better CPUs.

Intel would be well served to push IEEE-1394 (Firewire) as it encourages people to use their desktops for highly CPU intensive things like video editing.

Firewire isn't just for DV!

[coene (554338)]

I dont know why every time Firewire is brought up, someone mentions it in relation to DV.. DV is one of the simpler uses for Firewire, the real treasure is in its ability to link ALMOST ANYTHING!

This really is very cool stuff, and although I'm as suprised as everyone else about Oracle releasing open-source software (GPL nonetheless), it's another huge step forward.

Things like this piss off Microsoft to the Nth degree. That rocks!

Re:Firewire isn't just for DV!

[budgenator (254554)]

This really is very cool stuff, and although I'm as suprised as everyone else about Oracle releasing open-source software (GPL nonetheless), it's another huge step forward.

if memory serves me correctly oracle announced a while back that it (the company) was going to be running on Linux starting this spring. It make a lot of sense to release the lib's to make it happen back to the community because now companies are more likely to be able to afford oracle software due to reduced licensing costs on Linux. thinking of FireWire only as a DV bus is like thinking of SCSI as only a scanner bus.

Survival Tactics

[bovilexics (572096)]

Actually, this is of no surprise to many that have followed Oracle over the past few years (perhaps 5+).

Oracle has been incoporating many open standards into their products recently which has been necessary to help keep the company in a (relatively) good position in the database server market. In the past all of their technologies were proprietary with their custom SQL extensions and their custom language for stored procedures and triggers (PL/SQL). Oh, and Linux - forget about it.

However much of that has changed and now they support Linux, XML, Java (I believe the first to have Java stored procedures), and a large portion of the J2EE platform with things like OC4J (their java app server based on Orion).

See these links for just a sampling of what I'm talking about.

Java Stuff [oracle.com]
Linux Stuff [oracle.com]

Re:Survival Tactics

[bigmouth_strikes (224629)]

>help keep the company in a (relatively) good position in the database server market

I guess that with a market share of 54% [oracle.com], "relatively" is the key word in that sentence...

Re:Survival Tactics

[NineNine (235196)]

Well, I wouldn't call their position in the database market "relatively" good. They're #1, and have been for a loooong time.

The Java stuff is cute, but by and large hasn't been implemented much. People buy Oracle because it's been around forever, and has been tested probably more than any other software on the planet. PL/SQL is still, by far and away, much more popular than their Java app. PL/SQL is incredibly optimized and solid, whereas their Java solutions are still getting there.

Their XML parser is definitely good, but the documentation for it is virtually nonexistent.

I don't think that they're necessarily adapting because they have to. Their core business is very strong. I think that they're just trying to expand their market. Of course, they've had lots of misses too. Some of their apps, like Oracle Forms (which is incredible) and their very nice web server while used, aern't nearly as popular as their core RDBMS.

And you forgot one of their coolest new technologies... OODBMS. Very bizarre. Very different. Hasn't taken off yet, but I've used it, and it's very very innovative.

Oracle's not in any trouble *yet*. But I think that they're hurt every time they try to work their way into the low end market to compete against things like MySQL. Bad idea.

Re:Survival Tactics

[Cecil (37810)]

Um, OODBMS is not [exln.com] really [versant.com]
that [neologic.com] innovative [adb.com], although I will agree that it is cool. I prefer PostgreSQL [postgresql.org] myself, but that's because I don't have tens of thousands of dollars to spend on all the commercial databases. *shrugs*

I apologise in advance if Oracle has redefined OODBMS to mean something different than I'm used to it meaning, but at least as much as I know what it is, it's hardly innovative. It's been around a very long time.

Re:Survival Tactics

[Petronius (515525)]

Actually, Sybase had Java stored procedures first.

Great, now I have to...

[Ayanami Rei (621112)]

buy a whole new round of motherboards that are firewire enabled! I wonder if you can create ring configs if you have two roots per PC.

I wonder when Oracle is going to buy a company that produces firewire interface controllers... can you say instant SAN business?!?!

Just kidding, I think...

Red Hat to use Oracle's cluster software

[bstadil (7110)]

There is more Oracle news announced. This [infoworld.com] was just posted over at InfoWorld. Me thinks its as much a blocking move towards .NET, see below.

From article.

: Linux backers are working to strengthen the OS and bring it closer to competing with the proprietary versions of Unix that currently dominate the data center. Adding a clustered file system into Red Hat Linux is another step toward this larger goal.

hmm, not much there

[jpc (33615)]

After going through the crappy registration process, what do I find: not much at all.

The (code not available) firewire stuff is a fix to allow sharing of firewrire disks. Which has been in the kernel for quite some time (perhaps they submitted it), but it is hardly radical (couple of lines of code, if your hardware happens to support it).

Seems more like a PR announcement to me.

Shared Disk

[by Anonymous Coward]

Does anyone know how firewire makes it any easier to share a hard disk between systems, for clustering support? According to the Oracle description of the patch "Firewire allows developers to easily and cheaply build a clustered system on a shared disk, which is useful for testing clustered applications...".

In a normal cluster configuration, SCSI provides an interface for allowing a hard disk to be shared between actual servers, so that if one goes down another can take ownership of the SCSI disk. Fibre is a common carrier, linking the computer systems to a disk array system (SCSI over Fiber), and Firewire could be used to replace it, but is the only benefit its expense?

Proper way to connect these

[bill_mcgonigle (4333)]

A cool project like this ought to be interconnected with a Hubzilla [charismac.com].

Firewire is not an alien technology

[kenp2002 (545495)]

Really people it's just a high speed interface. Not really that much different that an old serial port, just faster. Do you ohh and ahh over the fact you can hook up "almost anything" to a serial port? Of course not. Firewire is no more, or less, versitle than USB, older serial, or even parallel ports.

Now is firewire had a liquid metal port that accepted any type of interface by morphing the connection, then firewire would be fucktacular! (Copyright 2003).

P.S. Starting throwing Copyright notifications on your posts, the "media" is starting to post OUR comments in their papers without our consent!

Firewire technology is important.

[phoenix_orb (469019)]

My friend, I am unsure if you are purposely being obtuse, or just don't know about firewire technology.

Firewire is hot-swappable. Try that with a external SCSI Drive. (not a hot swappable disk, the entire drive)

Firewire doesn't need a computer to work. USB 2.0 and 1.1 need a computer for it to work, but you can actually plug a DV camcorder straight into a digital VCR.

There is up to 50MB/s transfer rates (400Mbits/s) and the design is scalable, meaning the next iteration of Firewire will be 800Mbits/s, or possibly even 1.2Gbits/s

Ease of use: FireWire cables are a snap to
connectyou dont need device IDs, jumpers, DIP switches, screws, latches or
terminators.

Data and power: the FireWire cable carries data of course, but also power. I have one cable on my desktop for my iPod. It charges and synchs it to my iTunes with one wire. Serial doesn't do that.

USB 2.0 doesn't have real world speeds at the advertised 480MBs. Firewire does.

It is an industry standard. Bar none. Purchase a new digital 8 or mini DV camcorder. What do you get? A firewire port right on the side.

So basically, I wish all ports were designed with the expandibility of firewire in mind. I can do just about anything with it. Now even if I have a super-duper fast parallel port, there is tons of stuff I wouldn't want to do it with.

Re:Firewire technology is important.

[pmz (462998)]

Firewire is hot-swappable. Try that with a external SCSI Drive. (not a hot swappable disk, the entire drive)

Well, technically an external SCSI drive can be hot swapped (I've done it), but having the bus idle while swapping is important. It's just not a good idea to yank out a drive when the bus is active. I haven't worked with true hot-swap drives; my hope is that they leave behind an intact bus when they are removed (otherwise, I guess hot-swap would be pretty moot).

Re:Firewire is not an alien technology

[GlassHeart (579618)]

Firewire is no more, or less, versitle than USB, older serial, or even parallel ports.

The additional bandwidth itself brings versatility. Can you watch a movie, listen to Internet radio, or play a network game over a 9600 bps modem? Yes, technically. But you wouldn't, because it'd be painful.

Do you ohh and ahh over the fact you can hook up "almost anything" to a serial port?

No, because you can't. The classic serial port was already inappropriate for the bandwidth required of a printer, over ten years ago.

BUS Limitations

[ehiris (214677)]

I have a limitation on my Motherboard of 266 MBps due to the link from the north to south bridge.

Could you connect a firewire card on AGP so that you can make use of the full 400 MBps that Firewire provides?

Re:BUS Limitations

[ehiris (214677)]

Reply to self: 400 Mbps. The excitement is gone :(

Re:BUS Limitations

[afidel (530433)]

WRONG! your north to southbridge connections is 266MB/s (notice the big B) which is 2,128Mb/s which is roughly the speed of a fibrechannel connection (of course most fibrechannel adapters are installed on systems with much greater internal bandwidth). BTW, AGP is great for pushing data in one direction, but it sucks at bringing information to the system. For similar bandwidth to AGP get a server chipset based motherboard with PCI-X or infiband connections.

.. they're also in cahoots with mysql

[outsider007 (115534)]

I was over at mysql.com earlier and I noticed a large pop-up ad for Oracle 9i and I thought, hmm.. something fishy here, since when do companies advertise products for their competitors? I mean that would be like slashdot running microsoft ads.

oh wait..

Ahh but

[codepunk (167897)]

But does firewire support multipath io with load balancing? A single point of failure in the hardware is unnaceptable. On a more serious note this is great as it allows for developers to test on cheap gear. That san with fiber channel we just bought for our clusters was one expensive dog.

Oracle is being a GOOD Open Source Participant

[d3xt3r (527989)]

It's nice to see Oracle contributing patches, reference implementations, and useful sample code back to the open source community.

Oracle has jumped 100% on the Linux bandwagon and is pushing it as the OS of choice for RAC (real application clusters) and claimed to switch all their internal production servers to Linux in the near future.

To see them giving code and "lessons learned" information back to the open source community is awesome. This is the type of business and open source relationship that proiveds a win, win for both the commercial party and the open source parties involved. Oracle benefits from a free and stable platform while contributing back to that community code that can help make the product (Linux is this case) better for everyone else.

Thanks Oracle, nice to see you doing a good thing for open source.

Firewire for real clusters? I don't think so.

[tonyhill (590105)]

When I first read the post, I got pretty excited. Dreams of cheap clustering for scientific applications danced in my head. No more need for Myrinet [myricom.com], no Dolphin [dolphinics.com], just Firewire and Beowulf!

Then, I read some performance metrics on Firewire. High bandwidth. High latency [evaluation...eering.com]. Doh! The fairies stopped dancing for joy.

The problem is that in scientific computing, the time it takes for one node to say I need that data to another node, and actually get that data determines the performance of many more apps than does the speed of the CPUs.

So, until a cheap, low latency solution for communications comes by, real clusters will be communicating over Dolphin [dolphinics.com], Myrinet [myricom.com], or some other propietary technology [sgi.com].

Tony

Re:Firewire for real clusters? I don't think so.

[afidel (530433)]

Scientific clusters are not the only type of cluster you know. For instance Oracle Real Application Cluster is what Oracle thinks will be the companies future. These clusters may need the bandwidth of firewire et al and can live with the latencies.

Clueless Ellison

[L33t-Geek (614706)]

In reply to this Slashdot story larry ellison was quoted to say, "We did what? GPL? Open sorce? And what the hell is Firewire?" -Geek

GPL Nice... Registration bad.

[Alron (12242)]

A thing to note about this nice software from Oracle... from what I found, you have to REGISTER on their website just to get access to it... Registration requires everything... phone, company info, home address, company address, you name it. Kinda intrusive for a GPL thing, no?

Re:GPL Nice... Registration bad.

[FauxPasIII (75900)]

> Kinda intrusive for a GPL thing, no?

Then why don't you mirror it so the rest of us can download (and subsequently mirror) it without having to register ? The GPL guarantees you that right. =)

The cost of Oracle.... ummmm nothing really!

[IdleTime (561841)]

I have read several articles here why people prefer to use because Oracle is so expensive.

Unless you are planning to use it in a commercial setting, Oracle is free as in beer!

The latest version of Oracle for Linux can be downloaded from here [oracle.com]

No TCP/IP support

[heroine (1220)]

With all the shared filesystem, process management, localization features, they don't support the most basic of all: TCP/IP over firewire. Then again, we wouldn't be in a recession if managers were producing something useful.

great!

[jafac (1449)]

now if only some enterprising storage device manufacturer would make an actual firewire drive, instead of the typical bastardized IDE-with-a-Firewire-bridge crap they've been selling. . .

Re:Why

[DuBois (105200)]

I could ask the same about copper Fibre Channel cables.

My suspicion is that, in both cases, the answer is: they're not in as much demand as, say, Cat 5 Ethernet cables.

Re:Why

[Twirlip of the Mists (615030)]

I don't know about FireWire cables, but copper FC cables are actually really heavily shielded. At least, they're supposed to be. So it's natural that they'd cost more than ordinary DB9 serial cables.