Since hardware seems to be such a big problem, why do you need to run hugely expensive Sun (et. al) machines?

I build PC's for work, and just did an Athlon 3200 with 2gb DDR400 and 300GB storage (mirrored) for £500.

Ok, some of the complex stuff needs a 'real' machine, but surely non-critical (e.g. stats etc.) could be run from one of these? Also, apart from hosting a database, why does the generation of work units need to be done on some expensive Sun. Surely it wouldn't be difficult to buy a few supercharged desktop pc's for a few hundred quid and have them generate work units. If one falls over, then so what? Reboot it! All of a 3 min job. As long as the data is mirrored then I don't see why this would be a problem.

I was running SETI1 for ages on home PCs so leaving them crunching 24 hours a day doesn't make them break!

Probably a reason why this wouldn't work, but being simple I can't see why it couldn't be setup to work, and reduce the work on the main PC's.

> Since hardware seems to be such a big problem, why do you need to run hugely
> expensive Sun (et. al) machines?
>
> I build PC's for work, and just did an Athlon 3200 with 2gb DDR400 and 300GB
> storage (mirrored) for £500.

Oh, man - that's just not possible...

Sun boxes are MAJOR pieces of hardware - the scope of their performance would smoke any pc/mac.

'-~



"Nipple Free Since 08/24/04"

Well first of all I THINK sun donated some of the machines. They are listed as being a sponsor of the project. Also as dave pointed out, there are many performance categories in which your average PC just doesn't hold up. The major problem they have been having is I/O bottlenecks. The hard drives just cant read/write information quickly enough. As good as IDE drives are for home users, they simply do not support the sustained data transfer capacity needed here. The SATA-RAID based snap appliance was supposed to be the answer but I'm not sure what is going on with that now. There is more to these servers than a good CPU.


---------------------------------------
- A member of The Knights Who Say NI!
Possibly the best stats site in the universe:
http://boinc-kwsn.no-ip.info

> Surely it wouldn't be difficult to buy a few supercharged
> desktop pc's for a few hundred quid and have them generate work units. If one
> falls over, then so what? Reboot it!

I can't seem to find any way you can buy desktop pc's with 4GB of RAM and raid arrays for a few hundred quid anywhere? Please tell me where you got those ;)
(joke, not flame)

And as it is rightfully pointed out, many of the machines and parts are not bought but sponsored by companies. The budget for this kind of project is pretty tight you see...

One of the main reasons for the existance of machines like the Suns and up through the ranks to the mainframes of today is that a PC does not really have the I/O capacity in sheer bandwidth that those machines have.

Companies keep buying mainframes simply because of that fact. The modern PC, though much improved over past generations has limited internal bus structures to be allocated to I/O even within the internal components.

One of the reasons you see many places with thousands of PCs is that, parrticularly in the early days is a cheap way to buy capacity. Because web servers work in a world that is "stateless" each request can be handled by any PC with in the cluster. Over time the management of the hundreds to thousands of servers becomes such an issue of cost that replacing the thousands of PCs with a single computer becomes the smart thing to do.

In the case here, you can also see,as someone said, that the equipment is donated. And of course, Predictor@Home did go to Dell and to my knowledge is still waiting ... :)

In past threads we have discussed/analyzed some of the I/O issues and at this time are at the conclusion that the allocation of I/O between computers and the disks attached onto them can be part of the I/O issues. As they move things around, performance can be monitored and the "proper" placement of the files evolves to the most efficient configuration. We have a brand new architecture and this has to be the old trial and error until the configuration can be "proved"...
<p>
For BOINC Documentaion: Click Me!

Janus - go to www.Aria.co.uk :)

This didn't include OS as we put it on SuSE Linux, and so OS and other software was free.

Thought it wouldn't be that easy, but if its Disk IO that is the problem the SATA should be as quick as SCSI...

I can't see bandwidth being a problem for stats though, they may just run a little behind, but better than not at all?