Linux To Power NWS's Storm Prediction System

Mr. Plow writes "The National Weather Service is migrating to Linux-based IBM workstations and has purchased 900 IBM computers and 160 servers to do so." He includes links to coverage at Forbes (a Reuters wire service story there), Government Computer News, and Computerworld.

Comments

Of course...

[kinnell]

With windows all you see is blue

Re:Of course...

[Ske]

and with IBM it is BIG blue...

linux to power....

[mrsev]

... I use electricity for power. Should I change to this "linux". Maybe linux is more secure.

One Problem...

[JamesD_UK]

Who wants reports favourable to penguins?

Bravo

[enodev]

Now I'll now 185 seconds earlier that a flash just struck my home :-).

The previous systems contained Hewlett-Packard PA-RISC processors with clock speeds ranging from 100 to 120 MHz. Although those specifications aren't directly comparable to the 2.4-GHz clock speeds of the dual Intel Pentium 4 Xeon processors, the new workstations, which have 2G of RAM each, run the various AWIPS applications noticeably faster than their predecessors, Piercy said.

For instance, one AWIPS app that used to take more than 60 seconds to start up now loads in 18 seconds, he said.

That's a increase of about 48 times of computing power and the app just loads 3 times that fast? Something has to be seriously wrong with this setup!

Re:Bravo

[kinnell]

That's a increase of about 48 times of computing power and the app just loads 3 times that fast? Something has to be seriously wrong with this setup!

Startup times are affected more by disk access times than processor speed. Disk access times don't increase anywhere near as fast as processor clock frequencies.

Re:Bravo

[AmVidia HQ]

load times is a wrong benchmark to pick, it doesn't show proccessing power well. Load times are mostly disk-bound.

Load times more than 3x faster IS a LOT faster. And no doubt they will have many many times faster actual data crunching speed.

Interesting

[frankthechicken]

Interesting, according to this

Linux first appeared on the NWS scene in 1995 in the form of Slackware

And in comparison to their previous HP-UX systems:-

. . .the switch has helped increase data processing speed.

"The performance that we measure with our benchmark has increased by over 100% since we completed phase one of our Linux migration," happily boasts Piercy.

Re:Interesting

[4of12]

That's the key issue.

The purchase price of high performance RISC machines has always been O(US$10^4) and the acquisition cycle a period of several years.

While you can still get high performance RISC workstations for those prices, it doesn't make any sense when comparing price/performance.

In 1993, PC's significantly underperformed RISC machines, so paying 5-8 times the prices was worth it.

In 2003, PC's perform equitably to the best RISC workstations: you need a really good reason not get a really great Linux workstation for $6-8K instead of high-end RISC hardware for $20-100K.

So what's the forecast?

[trystanu]

Unsettled weather for the duration of the migration?

Are forecasts really getting better as computer power increases? I know that Metrology is always in need of more computer cycles to model the weather, but have forecasts actually improved with this power? Are there any success statistics around?

Re:So what's the forecast?

[SpaceLifeForm]

They keep saying that the 5 day forecast is just as in-accurate as before, but now, they know this fact with more accuracy than ever before.

Re:So what's the forecast?

[frankthechicken]

Well, since Edward Lorenz and his discovery of chaos, many of the current weather prediction systems are chaotic, and thus the forecast becomes less accurate over a longer time period.

Like many chaotic equations, the systems of weather patterns follow a `normal' path for a period of time before becoming chaotic. So this method is useful in predicting specific weather patterns for as long as 5 days in advance. After this point though the chaotic nature of the systems become apparent, with different scenario runs of similar data leading to completely different results.

So, more advanced computer power may lead to a better understanding of the chaotic equations which drive our weather and other environmental factors, though I for one remain sceptical, there are just too many factors to take into account for anything more concrete than the weather might possibly be . .

Re:So what's the forecast?

[Xenolith]

Well, to throw one stat out at you. In 1987, the NWS had a probability of detection score for tornadoes of 25%. This past year it was nearly 80%. Now given, part of that improvement is better trained employees and an excellent skywarn/spotter network. But yes, bigger and better computers are helping to improve the NWS mission.

We have are workstations already

[Xenolith]

I'm a NWS forecaster. We have our new workstations... they are pretty darn nifty. They are dual Xeons (2.4 ghz) running RedHat 7.2. These are much improved over the old HPs, which were J200 (overview). The bad news is that the database and application servers are still the old HPs. These are scheduled to be replaced in the near future as well.

Re:We have are workstations already

[girouette]

The computing context is similar where I work (Meteorological Service of Canada). There is a big difference between the forecaster's (technical) desktop and the clerical desktop. Our technical desktop never belonged to MS WIndows. Cost remains a concern, obviously, which is why we are moving away from proprietary Unix.

For what it's worth, our next-generation workstation is going to be java-based. (Joint effort with a handful of European countries led by Germany.) http://www.dwd.de/de/Technik/Projekte/NinJo/ (German language)

Dunno about the NWS, but for our clerical desktop to move away from Microsoft would be nothing short of miraculous.

Windows

[RevMike]

Of course... With windows all you see is blue

Of course, the NWS is using Linux to replace HP-UX workstations.

Linux replacing Unix : just standard evolution

[slb]

What would have been far more interesting is a Windows to Linux migration with the same improvement in performance and TCO.

Maybe this will help Linux's respectability

[Maestro4k]

Having someone big and respected like the NWS using Linux and announcing it publically should help Linux's respectability quotient in the US a lot. I'm glad to see this happening, both because it increases Linux's usage out there, and because it'll save us money (IIRC, the NWS is taxpayer funded), and probably lead to more accurate forecasts. It seems every time the computers they use get faster/more powerful, the forecasts get better. :)

Important to mention...

[Anonymous Coward]

That the migration is from HP UX boxes to Linux, not from Windows as some people might assume.

My weather man well, frog

[snkmoorthy]

My frog, actually croakes and stares at the open window, whenever it is going to rain, otherwise he just stares at the TV - although I have started noticing recently, that he doesn't care much about rain if Sex in the City is on.

--Sig here--

lies

[n0k14]

Almost every day I read about a department, agency, company or even country switching to linux. I rarely read about switching from linux back to windows. Where is this inconceivebly high implementation and training cost for linux that micosoft keeps whining about?

Why not using distributed.net...

[Ummite]

Why not using free systems like distributed.net, wich is 1) more powerfull and 2) free. People would be glad to help predict better weather, since actually they can't predict correctly 24h in advance ....

Re:powered by linux....

[Orp]

I, for one, give a shit. I am a meteorology professor who has been using Linux for over a decade. Having the NWS transition over to Linux means it is possible for me to run the same software that NWS operational forecasters are using. In fact, that's already possible - I have a copy of the AWIPS software on a CD-ROM sitting in my office, ready to install. This, in turn, opens the door for research possibilities and just as importantly, allows me to expose students to the kinds of software they will be using should they opt to work for the NWS (many do).

I visited the Grand Rapids NWS office a month ago and most of the workstations were already running Linux. The SOO (Science Operations Officer) seemed to be pretty happy with this. Why shouldn't he - hardware and software costs go down, machines are faster, and the OS is something that most scientists are using anyway.

Leigh Orf