What is recycle rate ?

[nuutti]

I mean that what percentage of full tests are recycled after 5 days?

Yours,

Nuutti

[MikeH]

While we're on this topic, can someone please confirm exactly what the test recycle rate is? Is it really five days? And the reporting of an interim result resets the count? I guess this seems reasonable for regular PC / Internet usage, but one of the PCs I have running SB is a laptop that is sometimes off the net for greater than five days at a time. Whole block tests are now taking about 2 days (it’s a P3-800). Let's take the following scenario - PC starts new block, I'm then travelling for one week and PC is disconnected from net (but switched on enough to have two days idle to spare). When PC is reconnected to net completed block is uploaded, but it's already been given out to someone else, and if they have a fast enough PC they'll have already finished it.

Something seems a bit wrong here!

Another question. In this type of scenario, if I'm off the net for five and a half days, when I return the completed block, does the user with the reassigned block somehow get informed? Or are they just wasting their time? While the test times are small, this type of issue isn't too big a deal, but it will be soon as the blocks take longer and longer to complete.

The GIMPS client (Prime 95) allows the user to specify a vacation period. Something like this would be good. It would also be nice to see something on the web site in the user area where I could see a bit more than "4 pending tests". It would be nice to see each of those k/n pares, and a time to expiry.

[Nuri]

MikeH you have some good points here. I'm sure similar features will be implemented as we test for higher n's.

On the recycle rate issue, AFAIK five days is the "no update from client" period. In other words, it's "ok" to send one block per five day. So, theoretically, one can continue testing a k,n pair for 5*block number times without being dropped (ie. 150 days max for a 30 block test as long as you test only one connect once every five days). Of course, it wouldn't be nice to return a two day test in 150 days, but it's theoretically possible.

Also, it seems like block sizes are determined based on constant cEMs, not constant ns. Forex, it was 10 blocks for n's aroud 1,330,000, 20 blocks for n's around 1,700,000, and 30 blocks for ns around 1,950,000. So, n per block decreases but time to test each block remains almost constant as we go up in ns.

Hope those helped some.

:)

[nuutti]

When saying recycle rate I meant that what percentage of
numbers f.g. between 2,000,000 -> 2,100,000
have expired and assigned to new searcher.

Lets assume that we have one million numbers and it will take 2 days to test a number. If number is not tested during five days it will be assigned to someone else. Recycle rate is 10 %.
Then
1. round 1,000,000 numbers have been assigned and 900,000 returned. After five days we move to round 2.
2. round 100,000 numbers have been assigned and 90,000 returned. After 10 days we move to round 3.
3. round 10,000 numbers have been assigned and 9,000
returned. After 15 days we move to round 4.
4. round 1,000 numbers have been assigned and 900
returned. After 20 days we move to round 5.
5. round 100 numbers have been assigned and 90
returned. After 25 days we move to round 6.
6. round 10 numbers have been assigned and 9
returned. After 30 days we move to round 7.
7. round 1 numbers have been assigned and 1
returned. After 32 work is done.

I was looking after this effect.

Yours,

Nuutti

[Nuri]

Right Nuutti, sorry, we were off topic somehow. Didn't want to bother you at all, just tried to help Mike on his post.


I understood what you meant, and I also wonder what the real figures are.

My "guess" is that it was below 2-3% until now (just a guess looking at the change in graphs etc., hopefully an intelligent one).

Well, just another note. Looking at the log file under config menu, I see that out of the 20 tests my client has been assigned so far, only one had an n value smaller than the then valid max n. "If" the distribution of my data is representative (?) of the whole set of tests, then the recycle rate should be around 5%.

[jjjjL]

The recycle rate is < 5 %.

In the past two weeks, it's around 2%.

Since a single block report resets the expire time, almost no tests expire. The tests that do expire rarely ever come back, so the repetition rate is less than 1%. And even the 1% of tests that do get duplicated are still useful as the residues can be compared to insure the client results are reliable.

The expire time will go up later but the 5 day limit still makes sense for now. It will probably go up to two weeks again when n > 3 million.

-Louie