updates on homepage - GO LOOK

[jjjjL]

Everyone hit up http://www.seventeenorbust.com/ for new updates.

Here's a repost w/ comments:

"9/22: Server upgraded, Expiration time increased

The server is now recording version information for all transmissions. I will now have a better idea of who is running the older clients and what OSes people are running the client on. After I have about a weeks worth of data I plan to email those who are running the older, slower, versions of the client. After that the server will start denying requests from the v0.9.0 (and lower) clients since they complete work 30x slower and use 99% of the project's bandwidth. Once those clients are upgraded to v0.9.2+ the server will be able to be moved to a server room for increased connectivity.

Now expiration time is based on the version you're running:

v0.9.0 -> 1 day
v0.9.2 -> 3 days
v0.9.4 -> 2 weeks
v0.9.6 -> 2 weeks

Rational is that caching clients are far more likely to return the result as time goes on. 2 weeks is long enough for even my P90 laptop to submit! :) If someone else has another metric you think I should be shooting for or just has a comment about the project, come to the Seventeen or Bust forum. I frequent the forum myself (name: jjjjL) and I take peoples feedback there very seriously. A lot of the changes I have made to the project have come directly from this forum, so if you want the chance to directly influence change, come by. "

I'll post some info on client distribution by OS as soon as I have some data. Eventually it will become part of the stats.

Also, notice I added a link directly to this forum from the site. :D Hopefully it increases the traffic here and improves the SB community with more constructive idea sharing.

One other small thing I did tonight was fix the graphs so the "cEMs/second" label on the vertical axis doesn't overlap the graph's range numbers. I hadn't fixed it before because Dave and I always joked that it would sure be funny when we had problems like the labels being covered because the stats got too big. The humor wore off so I just fixed it. I hope to have the same "problem" again soon when stats top 10 McEMs/sec. :)

Anyone have a comment on the expiration times I arbitrily made up? Any thoughts on a good time frame for letting people upgrade before i start locking out the v0.9.0 clients? I'm also considering a mailing to all users to let them know that there are improved clients out... what do you think? Would that bug people or be helpful? I'll likely wait on that email until I have a caching linux client or possibly until the next windows/linux client is finished. We shall see.


-Louie

[ColinT]

Arbitrary numbers work for me:) Once you have a list of guys running 090 and lower, you should email them. The rest of us should know by now:)

[alpha]

Originally posted by jjjjL

v0.9.0 -> 1 day
v0.9.2 -> 3 days
v0.9.4 -> 2 weeks
v0.9.6 -> 2 weeks


[Mon Sep 16 07:01:34 2002] got proth test from server (k=27653, n=1584537)
[Sun Sep 29 18:46:37 2002] completed proth test(k=27653, n=1584537): result 3

The above was with 0.9.6. That block hasn't shown on the stats. Did I miss the deadline?

I think the block may have never been sent, though. The client couldn't contact the server when it completed. I closed the client after a while and started it up later, using the server and port suggested by Alien88, at which point a new block was fetched.

Also the logging seems too verbose when the server is unreachable:

[Sun Sep 29 18:46:39 2002] couldn't report to server [report denied], retry in 300 secs [error: -3]
[Sun Sep 29 18:46:40 2002] couldn't report to server [report denied], retry in 299 secs [error: -3]
[Sun Sep 29 18:46:41 2002] couldn't report to server [report denied], retry in 298 secs [error: -3]
...all the way down to 0 secs.

Any thoughts?

[jjjjL]

Here's why:

You retrieved that block several days before I increased the expire time limits. Unfortunately, the server had already deleted your assignment of it. I looked up that block. It was completed several days ago by The_Man.

You happened to have gotten the block just barely long enough ago that my new expire times couldn't save you :(

Shouldn't happen again.

Also, don't use the port settings alien posted unless you are behind a firewall. if the default ones works there no reason to use the proxy since it just routes the results though one more computer before going to the exact same server.


-L

[alpha]

Originally posted by jjjjL

You happened to have gotten the block just barely long enough ago that my new expire times couldn't save you :(

Also, don't use the port settings alien posted unless you are behind a firewall. if the default ones works there no reason to use the proxy since it just routes the results though one more computer before going to the exact same server.


OK, thanks for the explanation :)

I am behind a firewall, and assumed that was why I was recieving the 'couldn't report to server' errors. After the server/port adjustment all was well.

[Alien88]

Louie,
It doesn't route them through another computer.. it's on the same one as the server.

[jjjjL]

Originally posted by Alien88
It doesn't route them through another computer.. it's on the same one as the server.

yeah, i should have said it routes them though another datapipe on the same computer. the point is, it's one more thing you probably don't need between you and the server.

-Louie

[jjjjL]

I just compiled a special version of 0.9.7 for my SMP computer and figured I should probably post it for those of you who have dually windoze machines like me.

Grab it from the homepage and run it as your second client... it uses different keys for caching so it can run along with a regular 0.9.7 client with no problems.

-L

[prokaryote]

:cool:

Thanks. Might be a couple of days, I'm in a bit of a race on another project. Will be back.

[prokaryote]

Works great. Thanks.

[jjjjL]

Everyone go to the site homepage, http://www.seventeenorbust.com/ . Teams are working so you guys can make a FreeDC team or anything else you want.

Also, the team stats are not yet linked but here's a direct link for you to peek at how things are shaping up:

http://www.seventeenorbust.com/stats/team.shtml

They are far from done, but all the data is saving so they will get better as soon as I finish the stats scripts.

I haven't decided how to assign already completed work to teams yet. I'll probably have people email me... DON'T DO IT YET THOUGH. ;)

There will be time to do it all in good time. For now, just download v0.9.8.

-Louie

[smh]

Louie,

How does the expiration work for the new client? Is this still 2 weeks, or is the time reset when you report a block is done?

Sander

[jjjjL]

the timestamp of your block is updated when you submit a block, so as long as it submits 1 block of the test every 2 weeks, it won't expire :).

it's actually really really nice that way. i can't wait until everyone starts using this version... it will be so much better.


-L

[Halon50]

:cheers: This is great news, and certainly worthy of a separate topic here on the forum! (Hint, hint) :D :cheers:

...But I still have a couple questions: How does it handle installing over the old client? Is it better to uninstall the previous version to clean up files first? Will installing 0.98 over 0.97 force the client to start crunching from 0%?

[jjjjL]

installing over will work fine. i think it will work if you uninstall too but i wouldn't bother. i always install over myself. nothing will be lost.

-L

[Halon50]

A little testing answered my own question: installing over the 0.97 version did NOT reset prior computation. :D

Only have one minor qualm left: Is there a way I can still get the client to reset its benchmark stats? I tend to leave the client running in the background when I go gaming, and it detects this as time it has 0% to 4% of the CPU, so the benchmark drops like a rock. Previously, stopping/starting or exiting and restarting the client fixed the benchmarks. Any chance of getting it resettable again, just to make me feel good? :)

EDIT: More testing has revealed that exiting/restarting the client resets stats. LOL, I really need to test more and yap less. Just too excited to see this release I suppose. :p :p

Going to go convert the farm over from SETI now. :cheers:

[smh]

Originally posted by jjjjL
the timestamp of your block is updated when you submit a block, so as long as it submits 1 block of the test every 2 weeks, it won't expire :).

Thats what i was looking for :|party|:

Do you plan to lower the 2 week expiration time a bit to a couple of days or 1 week?

[Mystwalker]

Hi there!

I just registered yesterday - so my first client update took place in less than 24 hours... :)
Solved some things I wanted to address (~55 hours needed for 1 WU, "rare" savepoint, ...). Very good.

So I have some minor suggestions (only cosmetically):

- How about a field which shows the time the whole n needs to complete (as it was the case in prior versions)?

- I think you can be so generous to spend the "blks done" another "oc" ;) The buttons at the bottom should limit the min width of the window.

- Just got the status "n.high = 18577 . 123 blocks left in test". To be honest, it took me a second or two to figure out what was meant. :eek:
Maybe
"highest n so far: 18577.
123 blocks left for this n"
or something like that would be easier to understand for newbies (like me ;) ).


And there are four final question:

1. AFAIR ver 0.97 took 4-6 MB RAM. Task Manager now tells me SB uses 13 MB. Did the memory requirement went up that much?

2. Supposed some client submits several blocks, but then stops transmitting within those two weeks. Does the next client client that gets that n start from the beginning or the first un-submitted block?

3. Could it even be practical to split high n's into several parts and let more than one client calculate it?

4. As there is no benchmarking results available yet:
What is the "best" system type for crunching? Important for me is the comparison Duron vs. P3-m...


Keep up that good work! :cheers:

[jjjjL]

smh - i'm gonna leave expiration time at 2 weeks for now. i'm hoping that expiring blocks out won't happen as much now that the retry code is improved. right now there aren't enough users to seriously tap out the server anyway so we're ok. as the stats show, there are rarely over 200 blocks pending. now, if that balloned to 600 or 1000 it'd be more important. on one hand, it's nice to have the lower blocks get finished fast so the lower bound can move up, but by expiring blocks too quick, there is a risk that work will just have to be redone, and there is no need to waste computing time just to force the lowerbound up.

Halon50 - glad to have you fully on board :)

Mystwalker - ok, n.high is a little confusing i suppose. if it isn't completely clear, n.high is how far in the current n you are.

as for spliting blocks across multiple users, that has been addressed in this forum thead
http://www.free-dc.org/forum/showthread.php3?s=&threadid=1685

SB is optimized for all processors well. intel chips are fast, AMD are faster yet. I've heard Xeon's are actually rather fast too so if you have a server you admin somewhere, it may want an SB client on it. ;)


anyway, i'm about to go to the Michigan v MSU game. maybe you'll see me on ESPN this afternoon ;)

later,
-L

[jjjjL]

btw, a quick way to see how long the full test will take is to close the client restart it, and then go in config and uncheck the "transmit intermediate blocks" checkbox and Save. the timer will then switch over to the full compute time of the entire test. make sure to switch it back though so your client will contact the server more often, which will impove stats and keep your block from ever having the chance of expiring.

-L

[MAD-ness]

I really like the new page design.

[Mystwalker]

@jjjjL

Yeah, saw you at the game - btw. what do you look like? :p
Naw, don't think that game was broadcasted here. Who won?


One quick note:
On my personal stats page the "cExpMods done in the last 2 weeks" value seems to be incorrect. As I'm on for 2.4 days, shouldn't it match the "total" value?

Oh, and how resource-intensive is it do request that page? I feel guilty visiting it several times a day... :D

[jjjjL]

We won.

BTW, i happened to see a few DC folks at the game.

A group of people sitting in front of me had tiny signs for team vodka martini -- part of another project that just finished.

anyway, i'll look into that stat thing, that doesn't make much sense does it.. hmm

i've noticed there is also a bug in the 24hour team stats since TPB is almost last when they should be #1.

-L