Debian Cluster Replaces Supercomputer For Weather Forecasting

wazza brings us a story about the Philippine government's weather service (PAGASA), which has recently used an eight-PC Debian cluster to replace an SGI supercomputer. The system processes data from local sources and the Global Telecommunication System, and it has reduced monthly operational costs by a factor of 20. Quoting: "'We tried several Linux flavours, including Red Hat, Mandrake, Fedora etc,' said Alan Pineda, head of ICT and flood forecasting at PAGASA. 'It doesn't make a dent in our budget; it's very negligible.' Pineda said PAGASA also wanted to implement a system which is very scalable. All of the equipment used for PICWIN's data gathering comes off-the-shelf, including laptops and mobile phones to transmit weather data such as temperature, humidity, rainfall, cloud formation and atmospheric pressure from field stations via SMS into PAGASA's central database."

One thing always missing from such stories...

What was the age and the specs of the SGI being replaced?

Going by Moore's law, a factor of 20 performance improvement takes about 6 to 8 years. If the SGI was at least that old, this isn't news -- it's just the state of the art these days. In other words, small clusters capable of weather forcasting are relatively run-of-the-mill.

Of course, props to linux for being the enabler in this case.

Re:One thing always missing from such stories...

6 to 8 years, you say? Well, then, they'll be ready to upgrade about the time the next version of Debian is released.

Re:One thing always missing from such stories...

This is inaccurate, as a long time Debian user I really cannot resist in correcting them.

> Debian "unstable" Sid is upgraded every day, or at least several times per week.

True.

> Debian "testing" is upgraded several times a month.

Wrong. Debian testing is updated automatically from packages in Debian unstable. The difference is simply that a package has to sit in Debian unstable for a few days, and no significant bugs can be introduced by the new package, before it is updated. Since the process is automatic, Debian testing is updated just slightly less continuously as unstable (it depends on the robot to check the package dependencies and bug reports rather than the maintainer to upload a new version).

The only time when the update rate is seen as low as less than that is when testing is in deep freeze, i.e., a new stable is about to be created.

> Debian "stable" is upgraded every one or two years.

It usually takes slightly longer than two years.

> The only one I have avoided is "Debian experimental"... :)

You cannot have a pure "Debian experimental" system. Debian experimental are subsystems that could have profound effect on the rest of the system, and so is provided for trial in isolation. E.g., back in the Gnome 1 to Gnome 2 transition days, or XFree 3 to XFree 4 days, these subsystems are tested in experimental before moving to unstable. These packages are supposed to be used on top of or to replace some unstable packages. Since they affects one particular subsystem, experienced testers can try one particular one based on their needs.

This isn't anything new...

Most weather prediction centers have adapted their weather forecast models to use Linux clusters. By running an operational forecast model on a cluster, it's easy for forecasters to scale the models so that they can be run (albeit slowly) on desktop machines, and are easily worked on by real meteorologists (versus IT professionals). At my university, we use a large cluster of machines on a RedHat enterprise system, and then able to scale the models and run them on multiple processors using MPICH compilers and batch jobs. Really, using a Debian cluster is no different then using a RedHat cluster. My colleague has access to the NOAA machine, which has more processors then you can shake a stick at... he talks about some code that takes 3 days to run on his personal workstation that takes 2 minutes on 40 processors. With the relatively low cost of a linux cluster, weather forecasting models can be run quickly and efficiently on numerous processors at a local level. With the ease of use of a Linux machine versus some of the supercomputers, it puts the power in the meteorologists to make those changes to the model so that it can improve forecasts.

Where is...

I tried:
apt-get -f -y install gweather
But it failed with something about "ldconfig: /lib/libearthquake-2.3.so.0 is not a symbolic link"

Is libearthquake in unstable?

Re:Where is...

You've got it all wrong; you should be using built-in tools like these:

more weather - For when you need a new update.

less weather - Got too much weather? Reduce it!

vi weather - When you want to change the weather.

emacs weather - When you want to change the weather on 15 separate planets at once.

cat weather - It's raining... oh, never mind.

Re:hmm

that would be HURD

Re:hmm

More importantly, is it Vista capable?

Yes, but only home basic.

Re:Debian?

i would suppose that debian is quite a versatile distro for any purpose...

-- from a debian user... who actually started quite late with potato....

Re:Debian?

Actually I don't like Debian much as a desktop machine, but I love it as a number cruncher OS. I've had a 10 machine openmosix cluster going for several years now, problem free.

Stability is a major thing with Debian, and my experience has been that this is quite true.

Re:Debian?

you shouldn't be running anything but compiled source on a performance cluster.

you wouldn't be running anything but source compiles on a performance cluster. :p

Re:Debian?

The binary package management really says it all.. you shouldn't be running anything but compiled source on a performance cluster.

Wow - how many performance clusters do you run again?

Not that I run a "performance cluster" as such - but I do run a bunch of machines that are very busy, all on Debian.

You know what? We compile the couple of programs where CPU is the bottleneck from source. We also compile Cyrus IMAP from source because we apply a pile of patches, but if someone else was packaging up all those patches in upstream, I'd be happy for them to be compiled there. Disk IO is the issue with Cyrus, and a custom compile won't help with that.

Yeah, we build our own kernels as well - that's another point that's worth the effort to customise. /bin/ls though? I don't think it matters to anyone on a high performance cluster. Just so long as the cluster apps are optimised then the rest is just noise - better to have a system that's less work for your administrators so they can concentrate on what's important.

Re:Debian?

Exactly, My job is running high performance computing clusters. You don't need to put much effort into your cluster distribution at all. As long as it's stable, and gets the job done, why mess with it.

The things I (and my co-workers) put a lot of optimization effort into is the kernel and our apps. You're exactly right.. 99.9% of our CPU cycles go into getting work done, and that 0.1% used by /bin/ls don't matter a bit.

Re:Debian?

Give the guy a break. He just wanted to see Gentoo compile in less than 2 weeks; he had a cluster handy...

Re:Debian?

Actually, Debian is intended for servers and runs on more architectures than any other distro. The whole reason for the long testing cycle on Debian is to make sure it's as stable as possible so it can sit on a server and need little or no attention for days, weeks, or even months at a time.

I hated Debian at first because it wasn't friendly, but I looked into it more and realized it was the best choice I could make for my production servers. I can set them up and check once a week or so and they're still chugging along without need of intervention.

I wouldn't use Debian on my desktop (I use Kubuntu), but it can't be beat for servers.

It's NOT a desktop distro. Especially compared to Mandr* or Ubuntu or many others out there.

Re:Debian?

Debian works out just fine for these kind of tasks. Here in the Netherlands the national compute cluster Lisa runs on Debian (http://www.sara.nl/userinfo/lisa/description/index.html) with 800~ to a 1000 nodes (I think the page needs updating by now).

Re:Debian?

Debian, like most distros, is what you want it to be. Debian is used regularly as a server OS.

I doubt they have X installed on these machines.

Re:Debian?

Debian is sure not a desktop distribution. Ubuntu would be one. I run Debian on all my servers.

Re:Debian?

Why Debian? A desktop distro? That's got to be one of the least scalable and cluster-friendly distros..

Keep going, I was expecting to hear you claim Windows XP was a server OS next...

Not sure why you call Debian a desktop distro, it's much more useful as a server.

Re:Debian?

You obviously doesn't have a clue about Debian. Debian has been a fine server since 1995, and I still choose it before RHEL anyday. I have always found it strange that everyone went for RedHat, when they could have Debian. Mark S. saw the advantages of Debian and based Ubuntu on it, Ubuntu is a server and a desktop distro, based on Debian. It has made more people realize the strength of the Debian approach.

Re:Debian?

Why Debian? A desktop distro? That's got to be one of the least scalable and cluster-friendly distros. If they would invest a little to set things up properly they could get a lot more performance out of their machines.

Debian isn't - and never has been - a desktop distro. If you want a desktop distro built on Debian architecture, you get Ubuntu, or Knoppix, or one of a dozen others. Debian's unique selling proposition is a combination of stability, which is very important to production servers, and a rigorous commitment to free software. Packages don't make it into Debian Stable until they have been thoroughly tested. Debian also has the best package management system in the industry.

Frankly, I wouldn't run a server with anything else.

Re:dent in the budget

Sure. Add in paying for tech support or the cost in man-hours it takes to keep it running. Both can make a serious dent where nobody expected to see one.

Re:I don't understand the difference

Many distro's add kernel patches and add different drivers to the initrd.
Also the core os ( most minimal installation ) has many different tools and libs.

Also at time of release they can pick from many different versions of a single package.
That in combination with what version of GCC and compile flags can and does make a huge differance.

And at least with Debian you really do know how the systems was build, with RedHat I still wonder...

Marcel

Re:I don't understand the difference

Because each major distro, while they use the same base kernel, GNU command line tools, and same GNOME/KDE environment, can have radically different kernel extensions and drivers implemented by one distro doing development but not another. If you're using whatever GUI tools a distro provides, they can each configure the same backend very differently, which depending on how the tool writes the config file can also effect stability, security, and other functions. Also Fedora/RHEL and tends to use tools created or modified by Red Hat specifically while those aren't easily available for Debian or SuSe, which have their own tools in the same manner.

Re:I don't understand the difference

Debian will run multiple services reliably under heavy load. From my limited experience, it's one of those distros where you "Set It And Forget It" and that's that.

Once you got it configured the way you want it, there's little intervention involved to maintaining it. It'll just keep chugging along. The keyword there is "correctly". Follow the readmes, howtos, and best practices, and you're golden.

It's also one of the oldest distributions which always kept to the spirit of GNU/Linux in general: community development and enrichment. Debian developers pride themselves on that spirit. To make the best software for humans. (At least that's what I gather from hanging out with Debian folk) These people are not only passionate in the software that they write, they do it without wanting anything in return, being humble in the way they do it, and wanting no reward for doing it. To them, their reward is in other people using their software and loving it! In my opinion they're not recognized enough.

But what do I know? I just use the software.