Guys,

I try to be as "nice" to WCG as possible in terms of returning work units on time, but I have an idea that could contribute a lot of compute power, but could break this model

I'm not expecting too many people to be experts on this, but Amazon Web Service hire out some very powerful computers for very little. During off peak time, any unused compute capabilities are effectively "auctioned" off. Typically you can get a 32 core Xeon machine for 30-35c in during off-peak hours . So if I got say 10 out of 24 hours, I could be donating a lot of compute power for $3-4/day. Peanuts. A small donation to a worthwhile cause.

The catch is, if you're "outbid", you lose your machine instantly - no warning, no graceful shutdown. And once the machine is shut down, it returns to it's pre-boot state, as if you booted off a read only CD.

Let's say we have a good scenario - I put in a bid for 35c. At 6pm, it goes quiet, and I win a machine. It downloads a chunk of work, and keeps working on it, returning work units until 6am the next morning. In that time, it might have returned say 320 work units, but there are 64 work units which it couldn't get to in time. Since the machine gets wiped, even if my machine restarts next evening, those 64 units are lost forever from my point of view - and WCG will wait for them to time out before sending them to someone else.

But that's a good scenario. On a bad scenario, I might get my server at 6pm. But then at 7pm, I get outbid by someone doing end of day reconcilliation. I send back 32 work units but lose 64. At 8pm I get the server again. But then at 9pm, someone else outbids me and although I've sent back 32 work units, and lose another 64.

It seems to me that WCG will then have a monstrous amount of work units it needs to wait to flagged as "no-response". Would this seriously mess up the way that WCG works? Am I better off not donating this computing time?

Thanks in advance for your response.