How long?

[AMD-USR_JL]

It looks to me like PrimeGrid is going to win the poll. Just to speed things (it takes longer to get everyone to vote than I thought) I went ahead and started the final poll.

[Bender10]

I almost think the 1 week sprint would be the way to go. Maybe start Saturday? And maybe take one of the other 2 projects, and start that 2 weeks later?

[gatekeeper53]

I voted for the 2 weeks because it is a pain in the rump to switch everything over and with 2 weeks we should be able to put a sizeable dent in any project.

[Evil-Dragon]

I'm game. Lets do it.

[Steve Lux]

I'm voting 2 weeks because I don't think 1 week is enough to make any real lasting changes in a particular project. Also, because it will take a few days for me just to clear out any remaining work units and get started on new projects.

[jmblazek]

PrimeGrid looks like it will be the pick. I like 2 weeks. Some things to consider:

- PG closes connections daily because of server load...therefore, make sure your cache is full to crunch through the downtime. Peak server load happens around 1600 UTC with usually a 4 hr runup and 4 hr rundown.

- Unfortunately, increasing your cache will lower your chance of being the first to return a prime. It doesn't happen a lot, but still something to consider. For those who don't know, a certain percentage of WU's are sent out for double check (initial replication 2)...the first to return a prime result gets the credit...for initial replication 1 WU's, there's nothing to worry about...it won't be sent out for double check until you return the prime result.

- Teams like BOINC@Australia have thought about doing PG but are waiting for the server to be upgraded. A big onslaught of users and hosts will definitely stress the server.

- Upgraded server is in the process of being bought, built, and installed.

[gamer007]

- Unfortunately, increasing your cache will lower your chance of being the first to return a prime. It doesn't happen a lot, but still something to consider. For those who don't know, a certain percentage of WU's are sent out for double check (initial replication 2)...the first to return a prime result gets the credit...for initial replication 1 WU's, there's nothing to worry about...it won't be sent out for double check until you return the prime result.

Never knew that. I always wanted to get as many WUs in my cache. I'll remember to do that once I finish up my last batch of NanoHive WUs.

BTW, when are we starting this? I'm ready whenever everyone is.

[jmblazek]

Never knew that. I always wanted to get as many WUs in my cache.

While it's not a big deal, most people don't know about it...that's why there's a small advantage to having a low cache...you get to beat the others to the punch. If everyone knew about this, then it would be back to a level playing field with everyone having "almost" equal chances at returning the prime first. fyi, the top prime finders know about it.

When I ran it, I kept my server connect below 0.1 days. I had at most 4 WU's in my cache at a time. Therefore, I'd run through my cache every 20 minutes or so (results are sent immediately, regardless of cache size, but if it takes you longer to get to them, it'll take you longer to return them). New WU's were downloaded after every 2 returns. Those who had large caches didn't stand a chance to return their results before I did.

However, if two people have their server connects at 0.1 days and they are unfortunate enough to get a WU with replication 2, then it's going to be the fastest computer that returns the result first and thus, if it's prime, gets credit for the find. The second person just gets the double check credit but no recognition in the Primes Database.

This is a bit unfair but it's the best PG can do right now. Looking at the surface of the inequity you'll see that doublechecking is relegated to the uninformed (those with large caches). If everyone had the same cache, then double checking would be relegated to those with the slower computers.

This inequity doesn't happen often...only on the WU's that are initial replication 2 and prime. I don't know what the current double check rate is but at one time it was 10%. So out of the 336 primes found so far for n=333,333 this has happened about 30 times since January. The only way to know exact numbers is to view the results, but that feature is turned off to keep server loads down.

Ironically, this little inequity was exposed on the first prime found at n=333,333. The WU was sent out with initial replication 2. The host that was sent the WU second was the host that returned it first...thus receiving the technical credit. Typical initial replication 2 WU's are sent out within a minute of each other.

Initial replication 1 WU's are not affected...PG waits till a prime is returned before sending the WU back out for double check.

So a short server connect time and a fast computer rules.

If you decide to minimize your cache, be sure to have work from another project to crunch while PG's server loads are high and not sending out new work. Otherwise, your machines will be idle until they can connect again.

Mind you, this is a selfish approach and not good for the ultimate goal of increasing team points. You should have at least a days worth of cache to maximize team points and to have enough WU's to crunch through down times.

Before the inequity was exposed, I used to have a couple days cache with a "use network only" between 0200 and 0800 UTC when the server experienced its lowest loads.

p.s. I haven't run it much since n=195,000 where I found 26 primes. So far at n=333,333 I have found 2 primes and credited for both. I don't think either one was an initial replication 2...fortunately!

[Evil-Dragon]

I already started work on it.

[gatekeeper53]

I have to ask. Is Primegrid really a good choice for a high useage project? Their server does seem to be a little unreliable and slow at best.