The New SOB forum

[ColinT]

Louie requested us to set up a forum space for him. To post here, please register. The process is quick and painless, or your money back.

Welcome!

[FoBoT]

:cool:

[Angus]

Can I post here?

I've been called an SOB a few times....




Seventeen or Bust sounds like some type of teenage angst thing...

[ColinT]

Yes, Angus, that's exactly what it is. I'll let you post once, but don;t make a habit of it :spank:

[Alien88]

*waves*

--
Mike Garrison
Alien88

[ColinT]

Hiya Alien88! I am SO glad to see all these people I have been competing against (And losing to:).

Welcome!

[jjjjL]

alpha - "It might be nice to have the progress percentage popup when you hover the pointer over the systray icon. "


i agree. i made that work last night. it will be in the next windows release for sure. also, i have found a way to implement colin's idea of a "time till completion" box. that will also be in the next version.

i've noticed a weird bug in the client too. for some reason, when the program boots, windows randomly sets SB's working directory to whatever the default working directory of the computer is. so if you close the client (not stop, but actually close it) and then reopen it, it may not find the cache file it's looking for, in which case it starts the test over from scratch. this isn't a dangerous bug, just annoying and inefficient... so if it seems like you loose your work, look around your hard drive for another copy of the log file (it moves that as well) and you should find your cache file... which you can move into the directory with sb, restart, and have it work. the cache has several validation checks too so don't worry about moving files around and corrupting data -- it won't let you. i realize this is a very ghetto fix... i'll make a real fix for it when i have time this weekend. in theory it won't happen too often but in practice i'm sure it's happening way too much.

oh, i added more details to the loging functions. messages now mention the specific n/k values under test. i may add % notes on the restart notices too. i think those changes would allow for rather interesting monitoring programs to be made for sb if someone was interested.

other plans for the near term include adding the ability to store multiple tests for offline processing. this will also make it easy to implement dual processor support.

i also found a way to make the cEM/s 1000x more accurate w/ less work than the current client uses. a few more changes and i may just start calling this v1. :)

keep making your suggestions. they're all really good.

-Louie

[ColinT]

Louie:

Could you explain the scoring system? When are the stats updated? Why are the graphs so ragged? Is one OS faster than another?

Things like this.

[ColinT]

Today I noticed that my output has been halved. It's taking 24 hours to do a WU where it took 12 before. WHat's up with that?

[jjjjL]

what are you basing this on?

your stats look about the same online. as the project progresses the tests will take longer and longer (until we swtich k values), but the way the stats work, it will compensate for the length of the test and the size of n.

think of the n size as the length of a test. plus larger n values take more than linearly more time. i. e. n=2,000,000 would take way more than 2x n=1,000,000.

at the begining i thought it would be nice to split individual tests into blocks to keep the time low.... however, then the blocks would have to be reported back to the server... and the blocks grow in size quickly. you may notice that the cache file is ~150k for the tests we're doing right now. do the math, and it just doesn't pay to transfer that up and back down 20-30x per test. each test then takes ~4MB of bandwidth. Multiply that out by a bunch of users and we're transfering multiple GIGs a month. :( since the server is technically hosted on a non-comercial line, we would be shut down so quick it would make everyones' heads spin.

compare that with the relatively bandwidth free model of the current server, and there is no way to argue in favor of blocks. if bandwidth was free, i'd do it.

this is also exactly why i knew caching would quickly become more than a nice feature, but a nessesity. if wu's become week long affairs, i'll switch the k value so we can have some instant gratification again but for now i don't plan to change the full-test model.

in conclusion... yes, tests are taking longer. cEM/s should stay relatively stable.

-Louie

[Alien88]

As the test progresses, the range is going to increase so a WU will take longer to do - however your cEMs/sec should stay roughly the same. I checked my logs and a few days ago the max N was around 900k, and now the blocks are around 1400k.

--
Mike Garrison
Alien88

EDIT: Of course louie posts right before i do..

[jjjjL]

all that said, i hope to make a new function that will allow the client to inform the server how far it along in the test so that the stats can be more real-time.

the long length of the blocks is what is leading to the jagged stats.

[Alien88]

I assume that'll mean for really slow computers that the server will know a computer is still working on a block and the time limits then won't apply to it?

[jjjjL]

correct. in fact i expect to extend the time limit soon when i'm convinced the majority of folks are using v.94 or better even before i add such a feature.

i doubt i need to make a big deal about it but i also plan to completely phase out the old clients (before v.92) whenever i get around to forcing the server to deny their logins. at this point, they are unlikely to be doing more than increasing the size of the blocks by creating bases in them.

-Louie

[Alien88]

Which brings up another idea.. in the GUI client(s) when a client is rejected because it is too old, have it pop up a window telling them their client is too old and to visit the web page to get the latest version.

[ColinT]

Yes. Distributed Computing changes algorithms every few days. They have a built in automatic client upgrade thingie which you can opt in for. The client checks for a new client every time it submits work to the server.

OK, I understand the nature of the problem, except my cEM/s are about half what they were, and it shows on my graph. My work is now worth less. If that makes any sense:) My P4 2000 is now at 28 hours into a block, and it's only 50% done. It used to do them in 12 hours. It's cEM/s are at 36000 and yesterday they were at 60000.

You two put on a good dog and pony show:) You guys look like an old married couple, finishing each others sentences:)

[jjjjL]

colin - so only your P4 took a hit? that could be understandable. the calculations probably don't fit into cache as well as when it was processing smaller numbers.

no one here needs me to dredge up a P4 v Athlon war here but there is no arguing that the Athlon has a superior FPU (even w/ SSE2 vectorization) and the Athlon has a larger more robust cache. plus the Athlon's cache isn't exclusive the way the P4s is. i'm in a comp arch class right now... we talk about these things all the time. the P4 is made to run at high clocks and to run small, computationally light applications fast... which are what most people actually do.

you gotta love AMD chips. all that said, i want a P4 anyway because i would love to have a 3GHz chip with some exotic cooling. :) i'll do it some time soon when i have more time/money. watch me.

-Louie

[ColinT]

Louie, you can have that 3GHZ chip this month, and all it needs is air cooling:)

I had to abandon my XP1800. It was on all day and did not advance past 29 hours.

The elapsed time indicator slows down and stops after 40 seconds. If I minimize the client and restore it starts going for another few seconds:) I am not sure what to do with the thing. It's alive but not processing.

My P4 is on 32 hours and only 64% done. This does not look good.

It's late. I'll get some sleep.

[Death]

bump

[poundsja]

In response to ColinT:

Could your halved performance be due to two copies of SB running? I've noticed the SB client tray icon disappears under XP. I would then start up another copy. Performance would be half of what it was. This bothered me one day (because it was one third of what it should be) and I decided to investigate. I brought up the process monitor and there were three copies of sb.exe running (but only one tray icon). I killed the running sb.exe processes and started sb again. Performance went back to normal. And yes, my stats went down (I think, which is why I decided to investigate).

[Alien88]

I dont know why this was bumped, i'm locking the thread.

[kugano]

I'd like to bring to everyone's attention the fact that this thread is almost a year and a half old. To avoid confusion, I'm closing the thread. (Actually, looks like Mike beat me to it.) Please start a new one.

Death, please don't do that again. Thanks.