Einstein@home gossip: Application ported to the PS3, but no SPE-support so far

Via the Einstein@home Cruncher’s Corner:

An Einstein@home cruncher ported the Einstein@home science application to Sony’s Playstation 3, but only using the Power PC core of the CPU. He did not use the SPEs from the Cell yet, which explains the floating-point performance as resulted on the Computer summary page for his PS3:

Operating System: Linux 2.6.16-20061110.ydl.2ps3
Measured floating point speed: 284.52 million ops/sec
Measured integer speed: 974.06 million ops/sec

Porting the science-application to support the SPEs of the Cell will be hard, without real compiler support. I can only guess what compiler he used and only speculate about the plans he has. Considering that the SPEs only do single-precission FLOPS he’ll have to find a way to implement double-precision in software (which isn’t a problem nowadays, algorithms exist). Also, the Sony CBEs don’t have all SPEs, it’s rumoured that only 6 out of 8 SPEs are utilized – my guess is that Sony gets the CPUs which failed the inital burn-in tests, the ones where some SPEs are dead. Cheap enough for a consumer-product.

However, interesting times ahead. Go, Gaurav, go!

About Einstein@home:
Einstein@Home is a program that uses your computer’s idle time to search for spinning neutron stars (also called pulsars) using data from the LIGO and GEO gravitational wave detectors. Einstein@Home is a World Year of Physics 2005 project supported by the American Physical Society (APS) and by a number of international organizations.

Tech Tags: BOINC einstein@home

BOINC: ‘Returning Results Immediately’ considered harmful

Via Romworld:

Rom Walton from the UCB BOINC-Team wrote an article about the impact of the “return results immediately” ("-return_results_immediately") setting of the BOINC-client. His point is that this setting puts a high and unnecessary load onto the project’s servers. He claims that leaving it up to BOINC when to send the results is about 70% more effective than sending them immediately because less database-queries are needed.

His figures make sense so i agree with him. So please don’t use that feature in your BOINC-client, especially considering the problems on some project’s servers in the past.

Thank you for your cooperation :-)

Tech Tags: BOINC

Rosetta: Article “Deciphering Protein Structures”

Via the NCSA:

The NCSA wrote a very easy to understand, yet quite complete article with explanations about David Baker’s Rosetta project, an theoretical approach to deduct a protein’s structure using computer-simulations.

Things I learned from this article:

1. The code does not start with a “flat” protein-molecule, starting to wiggle it around, but with a “homologous known protein structure” as a starting point. I don’t understand if that’s good or bad, but it limits the permutations to be checked.
2. David created a portal known as Robetta, where other biologists can submit their models to be crunched.
3. The Rosetta-project (not to be confused with Rosetta@home) uses a lot of CPU-hours on NCSA’s clusters and supercomputers (Tungsten Linux Cluster, NCSA Condor Flock, and now possibly TeraGrid resources)

However, quite a nice read, go and grab it while it’s hot!

Tech Tags: BOINC distributed computing Rosetta@home Science

BOINC: Why you should care about the credit-system

The probably heaviest request from users during the last BOINC user-survey was definitively “Introduce a more fair credit-system”. It’s still kind of frustrating that some projects hand out lots of credits per CPU-hour where others are more close-fisted with their credits. And there’s also the issue that we have “calibrating” BOINC-clients which sail around the known credit-issues and manipulate the claimed credits for a work-unit.

Some people consider this cheating, others claim that this is self-defence – their argument is “why should we get less credit in total even if we crunch more data per day?”

Both have a point, so some projects finally decided to go away from the naïve BOINC credit-scheme (which is based on the internal benchmarking algorithm) and create their own, CPU-hour based scheme.

For instance, Einstein@home recently “tuned” their credit-calibration (1) again to be more fair – silently – which caused an outcry from some people because they’ll issue less credits in general now. Rosetta@Home introduced a new credit-mechanism as well, but is more transparent (transparency is the major pro for Rosetta@Home anyway).

Why would someone who’s into science care about the credit-system anyway? There’re several reasons: Motivation and individual success is the absolute base for public voluntary distributed computing, something which some people out there didn’t understand yet. If you want to build up and maintain a large user base you need to give them incentives. Credits, public blessings, and – important – constant reports about the project’s success which show more than just how many percent of the project is already done like RC5 does. (OK, i have to admit, there isn’t much to report in the RC5-project, but you get my point, don’t you?)

And that’s the reason why you have to care about how many credits you issue and how you discuss the credit-issue in public – never underestimate the so-dubbed “credit-whores” – they’re your user-base and might wander of to projects which hand you more credits. If you’ve lost a user you’ll never get him back – most probably.

Be opportunistic and go for the high-performers even if they’re just after the credits. Be nice to your users and give them real reports every couple of weeks. Participate in the fori and give your users feedback. If possible, organize parties to meet your users (no one ever said anything about that you should pay). Optimize your science-application and be as fair as possible with the credits. Take rants and criticism serious. If people start optimizing your science-application: Embrace the changes and let them take part in the validation-process.

Remember: Even if tuning your science-application to be as efficient as possible takes a lot of effort, remember: Your users will thank you because they can crunch more data and you push your project onto a new level.

Corrections:

(1)

Bernd Machenschalk from the Einstein@Home-project correctly pointed out that they did not change the credit-system but only the calibration of the system they’ve introduced with the S5-run.

Tech Tags: distributed computing BOINC rants

Einstein@home S5 update

Ben Owen of the Einstein@home science-team posted an update about the ongoing efforts; this posting was done in the Science-forum, not on the frontpage, where one would expect it.

He reports that the National Science Board officially certified the project as “reached the initial design goal” which means “we’re officially in business”. He also points out that the S5 raw-data is twice as good as S4 was; they had some problems with the precission of their interferometers, mostly due to construction-works outside the L1-site.

Check out the forum for more details.

Tech Tags: einstein@home boinc distributed computing

RALPH@home: Alpha test project for Rosetta@home

Via rosetta@home’s Number Crunching forum:

RALPH@home is the official alpha test project for Rosetta@home. New application versions, work units, and updates in general will be tested here before being used for production. The goal for RALPH@home is to improve Rosetta@home.

So if you got any spare CPU-cycles and want to help improving cutting-edge rosetta@home applications, go and sign up!

Tech Tags: rosett@home boinc ralph@home distributed compunting

Rosetta@home: Return your CASP7-results asap

Today, on Monday the 7th, the CASP7-contest is over and the Rosetta@home-team needs to submit their results, as user Feet1st reminds us; so, if you’re a Rosetta@home-cruncher, return your result immediately through marking the rosetta@home application and pressing the “Update”-button in your BOINC-manager application:

Also, if you followed David Baker’s request for more more power, keep in mind that CASP7 will be partly re-evaluated and that further crunching helps science, although CASP7 is over.

Please keep also in mind that Rosetta@home got a 10-mio.-USD grant from the Melinda and Bill Gates foundation for searching a vaccine against the HI-virus. We don’t know yet when those new WUs will be issued, stay tuned for further updates.

Thanks you for crunching for Rosetta@home.

Tags:
rosetta@home, boinc, science, distributed computing, casp

New paper: “Designing a Runtime System for Volunteer Computing”

David P. Anderson, Carl Christensen and Bruce Allen wrote a paper about BOINC, the “Berkely Open Infrastructure For Network Computing”, a public Grid run by volunteers with a total peak-performance of about 350 TeraFLOPS.

The paper will be presented on SC06.

350 TeraFLOPS would be enough to challenge the Earth Simulator for the 10th place in Top500, if BOINC would be allowed to take part.

Tags:
boinc, supercomputer, grid, sc06

SETI@Home rumors hit the news

The Register reports thath there are rumors in the wild claiming the SETI@Home has finaly found an extraterrestrial signal.

Some guy called Steven Greer of CSETI, allegedly a a “professional SETI watcher” according to El Reg, alleged that SETI-insiders told him that they’ve found a really strong signal which was jammed from earth right after the discovery.

Sounds like yet-another-conspiracy and i find it quite amusing that those news hit the ‘net after people started questioning if SETI@home is the right thing.

Dear cruncher-colleagues from SETI@home: Don’t let those people ruin your day and keep on crunching!

Tags:
CSETI, seti@home, BOINC, rumor, conspiracy

New supercomputer: MDGrape-3 kicks some IBM ass

Website Newsfactor reveils some juicy details about the now to be operational new supercomputer MDGrape-3, developed by RIKEN, an japanese research-institute.

MDGrape-3’s main task will be protein-folding, pretty similar to what rosetta@home and folding@home are doing.

Interestingly MDGrape-3 is made of specialized CPUs, which are carefully handcrafted and designed especially for this purpose. Because of this MDGrape-3 takes the 1-Peta-FLOPS hurdle with just 4080 CPUs – and the whole system costs just a cheapo 9-millionen USD!
For comparision: The current No. 1 on the list of supercomputers is the BlueGene/L by IBM with 131072 CPUs, crunching for the Lawrence Livermore Lab in the US.

For a comparision i compiled some quick facts about MDGrape-3, BlueGene/L and the Earth Simulator:

	BlueGene/L	Earth Simulator	MDGrape-3
Vendor	IBM	NEC	Hitachi, Intel, SGI Japan (NEC)
Institute	Lawrence Livermore	ESC	Riken
Location	Livermore (CA), USA	Yokohama, Japan	Yokohama, Japan
FLOPS	280-Tera-FLOPS	40-Tera-FLOPS	1-Peta-FLOPS
CPUs	131072	5120	4808
Costs	250-300 Mio. USD (?)	62.4 Mio. USD	9 Mio. USD
Costs/GFlop	140 USD	8000 USD	15 USD
Power consumption	1.5 MW	18 MW	76 kW
Heat dissipation	1.6 MW	18 MW (?)	76 kW (?)
Footprint	232 square meter	3250 square meter	?

Yeah baby… boot up those FLOPS!

Update: Yeah, it’s FLOPS, not FLOP. I’d also like to point out that some people dount the 9-mio.-USD price-tag, considering that most of the workforce was provided by IBM and Hitachi, and NEC paid for most of the hardware. Got no more information, will keep updating on this.

[1] http://domino.watson.ibm.com/library/cyberdig.nsf/papers/D5D42E99C952A2CB85256FBA00755517/$File/rc23555.pdf
[2] http://www.llnl.gov/ASC/platforms/bluegenel/
[3] http://radio.weblogs.com/0105910/2003/10/01.html
[4] http://www.es.jamstec.go.jp/esc/eng/Hardware/system.html
[5] http://scape.cs.vt.edu/papers/sc04poster.pdf

Tags:
mdgrape-3, supercomputing, proteinfolding, bluegene, earthsimulator