Infrastructure Size

[the weaz]

How many more cpu's do you think that the infrastruture of the SoB servers could take on?

What would overload the servers?

[MAD-ness]

Very little bandwidth is used by the client to return results and fetch a new number to work on. The bandwidth usage is likely measured in kilobytes per day.


There is actually much more overhead due to Windows network layer transmitted than real data. If you run the client on a fast enough computer, you may burn a few kbs/day.

They recently (right before xmas) did a re-write of the server code and the stats code as well (I think). Because the amount of data being transmitted is so small and each proth test can take a long time (even with the client set up to transmit intermediate blocks it still doesn't report to the server all that frequently) the server load shouldn't be huge.

I have no idea what kind of load the server(s) could handle. That said, this project/client doesn't appear to be particularly bandwidth heavy, ala SETI, F@H, DF, etc.

Hopefully Alien88 will drop by and sum things up. He hosts the server I believe.

[the weaz]

say another 1000 cpus joined.

Are there enough work units? I see them doing sieveing and I notice there seems to be a range of k's being handed out.

Just idle thoughts. (maybe not but just checking ;):P)

[jjjjL]

+1000 cpus is nothing. in fact, i expect many more than that to join with virtually no impact. 10,000 cpus would be fine. 100,000 might require a processor upgrade for the server, but probably not.

it's academic anyway. you find a way for 100,000+ people to join, and i'll find a way to make it keep working ;)

the server is very elegently coded and can thread out to an arbitrarily large number of connections at once.

there are more than enough work units for the next several years, unless an insane amount of processing power came in (like 1 million cpus for a few months).

sieving just makes the workunits assigned more optimal. the more sieving done, the less workunits that need to be assigned but it won't reduce it very much.

as it is, there are over 700,000 proth tests online and ready. more could be created if those were depleted somehow. so in theory, 1 million cpus joining tomorrow would be too much... but if there were a month of warning before it got that high, we would be able to prepare more work.

-Louie

[the weaz]

ok thanks for the answer.

off to find a couple thousand cpus ;) :rotfl:

[MAD-ness]

Now all we have to do is find a way to get that many people to run the client and then keep them running it. ;)

BTW weaz, glad to see you guys trying out this project. Hopefully some of the firepower you bring over for the race with Anandtech will hang around afterwords. :)

[the weaz]

We'll see. TNT has kinda been running around like a chicken with it's head cutoff recently (aka Post ECCp-109).

We really thought TSC was the money project, but Sengent let us down big time. SoB has a good shot at being our big/main project along with DF. Well that is until something better comes along. :)

Who knows maybe after than TA race, we'll decide we want the #1 spot in SoB :)

[garo]

Ack! 700k Proth test in the under 3 mill range??? I thought we haev 746k tests for the 3-20M range due to our sieving.

Wasn't the 3M range sieved to 1T? Wouldn't that have reduced the number of proth tests? Inquiring minds want to know.

[Nuri]

Originally posted by garo
Ack! 700k Proth test in the under 3 mill range??? I thought we haev 746k tests for the 3-20M range due to our sieving.

Wasn't the 3M range sieved to 1T? Wouldn't that have reduced the number of proth tests? Inquiring minds want to know.

I think Louie meant from here (2.6 million to 20 million)

I "guess" there are only 17,000 tests left from 2.6 to 3.0 million.

[jjjjL]

correct. my number includes all the tests up to 20 million.

even though we are still actively sieving them (and will continue to), there is nothing stopping the server from assigning the tests. in other words, the tests are online. each time a factor is submitted, a test that would have otherwise been assigned is flagged in the database as being composite and so the server never assigns it.

well, there is something stopping the work currently being sieved from being assigned... and that's because the server always assigns the workunit with the lowest n value first. so sieving still has time before even the first value being sieved goes out.

it's not like i'm just collecting all the factors in a text file and will later add the work-units to the database by hand. they're in the database. the factor submission page works in real time with the actual server.

so each time a new factor is submitted, the total amount of work is immediately reduced. pretty neat, huh? :)

-Louie

[garo]

Yeah that's cool! Another thought came to my mind. Since the sieving efficiency has improved dramatically in the past week or so it may make sense to sieve say from 2.6 to 3 million as well. I remember - maybe incorrectly - that n <3M was sieved to 1T or was it 10T?

Anyway just a thought since factors are flagged in real time.

[jjjjL]

garo - good idea. there's still so much work to do in the n > 3 million range so for the moment, i'll do this resieving with a few of my own computers. it may reduce a little work. it's such a small range, the efficiency of SoBSieve really isn't leveraged like it could be so I wouldn't recommend it for most people. it will be difficult to judge when the efficiency is gone and it's also peskier since people may be getting assigned these values as the sieve runs.

however, the flip side is this range is so small.... SoBSieve runs very fast in p/s even if it's not as efficient as checking a larger range... it's neat seeing SoBSieve running at 700k p/sec ;)



-Louie

[garo]

Cool! Yeah it's probably best for you to do it yourself. Remember a PRP test saved is a PRP test done :)

[MikeH]

Louie,

That's almost (but not quite) answer to my question in this (http://www.free-dc.org/forum/showthread.php?s=&threadid=2424) thred.

I'm still interested to know how deeply the n < 3M range were sieved.:rotfl:

[ceselb]

Afaik I did the last range (1500-1600G) on 4847 (as seen here (http://www.free-dc.org/forum/showthread.php?s=&threadid=2058&pagenumber=4http://www.free-dc.org/forum/showthread.php?s=&threadid=2058&pagenumber=4)).
I have no idea how far the other were sieved though.