SB Linux client v097 being tested

[jjjjL]

the new caching linux client is finished :D

i'm testing it out now. if everything looks good, expect a new linux version this weekend.

FreeBSD will follow shortly after that.

BeOS may not have time... it's my b-day this weekend so I want to get away from my computer some. ;)

-L

[ColinT]

Well have a Happy One!:|party|:

[jjjjL]

Linux client done. FreeBSD binary also online now.

Everyone upgrade. Caching + verification are the two most important benefits. Both are very important for the project as a whole.

Now that there are new clients for all 3 main OS releases, I'm going to consider blocking the older versions. I will block v0.9.2 as soon there is a new windows client that can allow SMP.

-L

[igor]

With the new client, SMP speedup factor is 50% on both Linux and FreeBSD P3 machines. Just to be on the same page as everyone else, here's I calculated it. 1 iteration takes 12min with 1 job running, 16min with 2 jobs running. So in 48min, I can expect 4 iterations done with 1 job running, and 3*2 interations done with 2 jobs running. Hence, 50%.

Also I have a question about the verification feature which is advertised as new for Unix 0.9.7 versions. Who was responsible for verification part in older versions, the server?

Igor

[smh]

Also I have a question about the verification feature which is advertised as new for Unix 0.9.7 versions. Who was responsible for verification part in older versions, the server?

As far as i know the verification feature only means that the last 64? bits of the final result are stored.

To verify if a result is correct, the number should be retested with another program. If the final result is the same (if the last 64 bits are the same we can be fairly sure the whole result is the same) the test can be considered correct. If not, another test should be done until 2 matching checksums are found.

IIRC in the GIMPS (http://www.mersenne.org) project about 2% of all the returned results are incorrect. The bigger the number gets, the longer it takes to test it, the higher the chance that one of the itterations went bad (and thus the whole test)

What kind of error checking is done during a test by the program?
Does this project plan to verify numbers?

[jjjjL]

the client -
error checking is done every 128th iteration of the test, for all 50 final iterations, and another check is performed before each cache save.

the server -
there are actually several methods the server uses to verify your result appears ligitimate.


right now i am not duplicating tests. i will at least wait until i have insured the client is optimized as much as possible before i start dividing resources to verify results. i can wait until the client is "finished" and the user base is larger.

-L