SB v1.10 released

[jjjjL]

I just finished packaging the new SB client.

Here are the new features (listed in the *newly included* README.TXT):

-P4 SSE2 code improvements -- 10% faster
-test abandoning code removed
-multiple open socket bug fixed
-removed blocking network calls (i.e. occasional hangs @ 100%)
-more verbose logging on intermediate block transmissions
-registry hook for Mathguy's service added (sb-mg.exe no longer needed)
-Mathguy's service installer (v1.6) now packaged with main distribution
-can now finish test without downloading new one if desired
-new registry hooks for future add-on features (queuing)
-in-client signup interface improved
-no longer asks users to clear cache when entering first username
-uninstall fuction actually works now


These versions of Mathguy's service included with this version of SB are newer than any publicly availible versions right now. They also only need sb.exe so there will not be another sb-mg.exe file needed.

I just fixed the uninstaller a few minutes ago when I realized it didn't work. This installer is built with a newer version of NSIS so it's smarter now.

I think this version successfully fixes all major outstanding bugs and almost all minor ones as well. I'd recommend everyone upgrade immediately.

BTW, only 2 people have tested the P4 speed improvement. Mike said it was faster. Mathguy said he didn't think it was different. Everyone who installs this on a P4 needs to tell post here how it runs.

Download away:
http://www-personal.engin.umich.edu/~lhelm/sb110.exe

-Louie

[MathGuy]

Incidentally, I might mention that, although my P4's don't run any faster with the new version, the non-P4's I've run it on *do* by about 15-20%...

Go figure!

[Jammy]

Hey MathGuy . . .I just installed the newest SOB build and it does not show up in my systray any longer! I like it in my systray so I can check my progress during the day!

I tried using that code I entered in the Command Prompt but it still does not shw up!

Jammy :(



Originally posted by MathGuy
Incidentally, I might mention that, although my P4's don't run any faster with the new version, the non-P4's I've run it on *do* by about 15-20%...

Go figure!

[MathGuy]

Was the service stopped before you installed the new build? If not, that may have been the problem. Stop the service with

net stop "Seventeen or Bust service"

then run sb110.exe again to install everything, press Exit on the client that pops up, then run

net start "Seventeen or Bust service"

to restart the service.

If that still doesn't do it, post the last few lines of your sobsvc.log file here and I'll try to see what's going on.

[MathGuy]

Actually, I should probably mention that for everybody:

Stop the service BEFORE you run sb110.exe to install.

If you are currently using the service, install sb110.exe as follows:

Step 1 (NT/2K/XP): use the Service Control Manager, or run

net stop "Seventeen or Bust service"

at a command prompt.


Step 1 (9x/Me): at a Command Prompt, run

cd "\program files\sb"
sobsvc9x -q

Step 2 (everybody): run sb110.exe to install the new client, press Exit when it pops up.


Step 3 (NT/2K/XP): Use Service Control Manager to restart, or

net start "Seventeen or Bust service"

Step 3 (9x/Me): run

cd "\program files\sb"
sobsvc9x

[Jammy]

Yes I stopped the client before I installed the newest version. First it was hidden then when I rebooted the box (XP 1900) it still did not show up. I clicked on my bat file (Start SOB) on the desktop andd it started up visibly in the systray, but when I clicked on exit on the GUI it just exited and did not restart itself.

I will reinstall it tomorrow . . .I am too tired now to do it.

Jammy

[soni]

Hi jjjjL

Would it be possible to remove the registry part.
Its kind of anoying to have the install in the registry, when you could place it in a file in the folder.

The reason is, that it wont be possible to image a system with SOB installed, since the current workunit is mentioned in the registry.

So basically, when ever I try to re-image my system, I have the problem, that it looks after the old wu.
It is also impossible to move the sob install, without uninstalling/installing it.

Thanks for your hard work..

[DSheets]

I have installed the new client on some of my PCs at home and all went well. One P4 (2.8 DDR) got about 5% boost (297K to 312K), another (2.53 RDRAM) got about a 7% boost (440K to 470K). The P3 (800 SDRAM) got a 1% boost (115K to 116K) and the Athlon (2500+ Barton Dual Channel DDR) got a 1% boost (328K to 331K).

Hope that helps.

Dan:cool:

[Frodo42]

My PIII 450 Mhz improved something around 5%, the new features is just great hopefully this will help recruit/bring back processor power to this project.
Cant wait for the Linux version ...

[Beyond]

/me waits for the BeOS version...

[dmbrubac]

Mathguy

What are 'NormalizeXP' 'WUQueue Size' and 'Autodial Type'? I didn't see them mentioned in the updated docs.

Dave B

[dmbrubac]

I am going to add to Mathguys instructions for NT/2K/XP users (maybe for 9x - I don't know cuz I don't have any)

Step 0, but only if you've used set the restart properties sobsvc -m

use sobsvc -m again so the client and service doesn't restart before you get the new one in.

Step 4

reinvoke sobsvc -m

Edited a bit later:

No that isn't my problem. I am getting an access denied when installing the new sobsvc.exe. I've issued an sobsvc -u then restarted which fixed the problem. When I tried to delete it manually, explorer reported it in use.

By the way, all this is happening over a terminal services admin connection, so the problem may be specific to remote desktop. I hope it isn't because I'm about to U/G a machine across town.....

Dave B

[MathGuy]

Originally posted by dmbrubac
Mathguy

What are 'NormalizeXP' 'WUQueue Size' and 'Autodial Type'? I didn't see them mentioned in the updated docs.

Dave B

That's because they don't do anything yet! Well, that's not entirely correct - there are switches that will change those states and they will be remembered the next time they are printed out...not terribly useful at the moment.

Here is what they are going to be: "WUQueue Size" will be the size of a work unit queue that is maintained locally. When a client finishes an entire block, it will go get the next block from the queue instead of getting one from the server. Queue maintenance (sending completed blocks, retrieving new blocks) will then be done only when the machine is connected to the 'net (principally useful for dial-up). I also have in mind a sneaker-netting feature in which you can export unfinished blocks to other machines and import completed blocks from them. These would go from/to the queue and be maintained along with the other blocks.

"Autodial Type" will determine when, if ever, the queue maintenance routine connects to the 'net on its on. Possible types are "never," "when there is a completed block in the queue," and "when the queue is completely full."

"NormalizeXP" is a switch that will cause Windows XP to be "normalized" when the service starts. XP has a number of scheduling quirks that cause things to run more slowly than expected after restarts and at other random times...this switch will attempt to force all of that stuff to happen at once (at system startup) so that the clients run at a more even predictable level thereafter.

Those are the plans, anyway...possibly also handshaking for a monitoring utility that can monitor the output for a number of machines which are all running the service handler.

Unfortunately (fortunately?) I'm headed to Japan for a couple of weeks, so it will be a while before any significant effort is put into this.

[MathGuy]

Originally posted by dmbrubac
...I am getting an access denied when installing the new sobsvc.exe. I've issued an sobsvc -u then restarted which fixed the problem. When I tried to delete it manually, explorer reported it in use.

By the way, all this is happening over a terminal services admin connection, so the problem may be specific to remote desktop. I hope it isn't because I'm about to U/G a machine across town.....



Somehow or other the service was still running when the new sobsvc.exe was installed. The 'net stop "Seventeen or Bust service"' command should stop it and enable the new sobsvc.exe to replace the old one. You really shouldn't need to uninstall in order to upgrade.

It is possible, however, that since you are administering remotely that you need to issue

sc <computer name> stop "Seventeen or Bust service"

instead of the 'net stop' command which runs locally

[dmbrubac]

I'm headed to Japan for a couple of weeks

Mmmmm. Unagi. Yellow tail. $100 Watermelon.

Thanks Mathguy. When I U/G'd the machine across town the service definitely wasn't running, but it still wouldn't let go. I uninstalled the service, installed SB 1.1.0 into a new directory (moving Y and Z files, etc) then reinstalled the service. Happy Happy.

I will delete the old sobsvc after the next maintenance reboot.

By the way, I have a couple of 'lost' proth tests that haven't expired yet and I still have the Z files. Do you think it is possible to resurect them? I wouldn't have even tried prior to the 'don't get another block when done' option.

I hoped I could force an additional client and give it the lost block info. Any ideas?

Dave B

[eatmadustch]

yep, copy the zXXXXXXX file across and also the HKLM\SOFTWARE\LhDn\sob\cache in the registry. I don't know if it works (as I've never tried it), but I heard it works in this forum a while ago ...

[MathGuy]

Originally posted by dmbrubac


Thanks Mathguy. When I U/G'd the machine across town the service definitely wasn't running, but it still wouldn't let go. I uninstalled the service, installed SB 1.1.0 into a new directory (moving Y and Z files, etc) then reinstalled the service. Happy Happy.



Very strange...haven't seen that on any of the upgrades I've done (and I did a bunch while testing this puppy). Didn't do them via TS Admin, though.


By the way, I have a couple of 'lost' proth tests that haven't expired yet and I still have the Z files. Do you think it is possible to resurect them? I wouldn't have even tried prior to the 'don't get another block when done' option.

I hoped I could force an additional client and give it the lost block info. Any ideas?


You know, I bet you can...this is hopefully going to be automated soon, but I *think* this should work (assumes that the machine currently has a single client instance running):

1. copy the reg key "HKLM\Software\LhDn\SoB\cache" to "HKLM\Software\LhDn\SoB\cache2"
2. change the "k" and "n" values in the new key to be the "k" and "n" values for the "lost" test
3. stop the service
4. run "sobsvc -p:2" to add a second client. Also turn *off* all the restarts that you currently have on that box (-r, -k, -m, -s).
5. restart the service
6. On the (second) client that *should* now be running the lost block, run the config and check the "on completion, exit..." box
7. whenever the lost block finishes, stop the service again, run "sobsvc -p:1" and put whatever restarts you want back on (they really shouldn't be necessary now, though) and restart the service.

This is unnecessarily complicated at the moment because the "on completion, exit..." flag is global *and* I don't really know what will happen when the handler immediately restarts a client that stopped because of that flag. Alas, that testing is all in the "post-Japan" timeframe...I suspect that the client will grab a new block at that point, which is *not* what you want!

Another approach that might be easier would be:

1. stop the service
2. turn off all the restarts
3. *change* the "k" and "n" values in the "HKLM\Software\LhDn\sob\cache" key to reflect the lost block
4. restart the service
5. run config and check the box
6. when the block finishes, stop the service, edit "k" and "n" again to put them back to the current block.
7. put restarts back on and start the service

This will work, assuming that it doesn't take more than 10 days for the box to finish the "lost" test.

[dmbrubac]

Excellent!

I don't think there will be a problem with finishing in time. One test is at 99% and the other is at 72% I will be running on a P4 that does 300k-ish.

By the way, there shouldn't be any reason why I can't finish more than one block simultaneously, other than performance oss due to context switching, etc. right?

Dave B

[MathGuy]

Originally posted by dmbrubac


By the way, there shouldn't be any reason why I can't finish more than one block simultaneously, other than performance loss due to context switching, etc. right?

Dave B

Correct - in fact, the active code of two clients should fit in the cache comfortably so combined throughput shouldn't drop much at all (there's already a fair amount of context switching going on with general OS nonsense).

[dmbrubac]

Alas the Z files are now gone. Perhaps the new client is better at cleaning house.

I'll bet they were primes too. :(

[MathGuy]

Originally posted by dmbrubac
Alas the Z files are now gone. Perhaps the new client is better at cleaning house.

I'll bet they were primes too. :(

Without a doubt!

Something like "the probability of a piece of toast landing jelly-side-down is directly proportional to the cost of the carpet"

[Jammy]

OK . . .I followed your steps:

I closed my SOB by clicking on the Stop SOB shortcut on my desktop. Then I went to Administrative Tools>Component Services anf stopped SOB as a service.

I installed the newest 1.1.0 and then I pressed exit and only the service started up and no icon in the systray again. I had to click on my Start SOB shortcut on my desktop to open up SOB so it would show in the systray.

OK . . .I did try inserting the script you emailed me last month but it seems not to be working with this new version. Mostly I like the fact that the script automatically restarted the client at 2 am for me.

Jammy



Originally posted by MathGuy
Was the service stopped before you installed the new build? If not, that may have been the problem. Stop the service with

net stop "Seventeen or Bust service"

then run sb110.exe again to install everything, press Exit on the client that pops up, then run

net start "Seventeen or Bust service"

to restart the service.

If that still doesn't do it, post the last few lines of your sobsvc.log file here and I'll try to see what's going on.

[MathGuy]

Jammy - the problem is that many of the switches are toggles - if you do them again, they undo what was done the last time. Post the last few lines of your sobsvc.log file (or e-mail them to me) and I can tell you what to do to get things fixed up.

[Jammy]

MathGuy . . .I just emailed you the sobsvc.log file from my XP 1900.

Jammy




Originally posted by MathGuy
Jammy - the problem is that many of the switches are toggles - if you do them again, they undo what was done the last time. Post the last few lines of your sobsvc.log file (or e-mail them to me) and I can tell you what to do to get things fixed up.

[priwo]

on my P4 Celeron sb110 is 2.5% faster than sb10

[tqft]

don't know what you did but my PII 450MHz now topping out around 60.559kCeM/s - previously topped out at about 57k CeM/s - about 5%.

Is there any benefit to running a second client (Win98?)? It looks like the client is using 100% but the SystemMonitor is not the best tool or very informative (ie not very granular).

Now if I can just keep Win98 from crashing and get some serious uptime and yes I do have a variety of linux distro's lying around but EAW will not work under any of them with WINE).

[kfx]

I am running a client on a computer that is not online all the time, I am using a dial up modem, so my problem is, that I have to watch the client closely, so that I can get on the net right b4 it is finish with a block. Could it be possible in the future to have an option like "send all blocks in queue"?

[Xeltrix]

Originally posted by kfx
I am running a client on a computer that is not online all the time, I am using a dial up modem, so my problem is, that I have to watch the client closely, so that I can get on the net right b4 it is finish with a block. Could it be possible in the future to have an option like "send all blocks in queue"?

It's already there. In config uncheck the box that says "transmit intermediate blocks". This way all data will be sent at end of test.

[kfx]

Thanks for the answer, but that is not what I am looking for. The client is running on a slow machine, so I can not wait untill the test is finish b4 I send it, at that time it will have expired.

[eatmadustch]

Originally posted by kfx
Thanks for the answer, but that is not what I am looking for. The client is running on a slow machine, so I can not wait untill the test is finish b4 I send it, at that time it will have expired.

it doesn't really matter if you're online or not when it finishes a block. If you're not online it will just send the block you didn't send with the next block, so you don't "lose" any work in the stats if you're not online when it finishes a block. You just need to transmit one block within ten days so your test doesn't expire, so I don't really see your problem!

[alpha]

Originally posted by kfx
I am running a client on a computer that is not online all the time, I am using a dial up modem, so my problem is, that I have to watch the client closely, so that I can get on the net right b4 it is finish with a block. Could it be possible in the future to have an option like "send all blocks in queue"?

I would also appreciate this. I find myself dialling up right before a block finishes so I can flush anything buffered. If I dialup too late, I have to sit there waiting for the next block to finish. A button to transmit everything done immediately would be real nice. Right after a new client release isn't really the best time for feature requests though, I guess :)

[kfx]

Originally posted by alpha
I would also appreciate this. I find myself dialling up right before a block finishes so I can flush anything buffered. If I dialup too late, I have to sit there waiting for the next block to finish. A button to transmit everything done immediately would be real nice. Right after a new client release isn't really the best time for feature requests though, I guess :)

That is the problem I am talking about, but I know that this is a very bad time asking for it. But this must be a problem for all people with a slow PC and a dial up connection.

[MathGuy]

I've been requested to post these instructions (which are probably clearer than my earlier ones) for installing the new client in such a way as to retain all the settings that you had for the previous service handler.

Here's the procedure (assumes that SB is installed in c:\program files\sb and that the upgrade installer is located there as well):

*NT/2K/XP*
Command Prompt -
net stop "Seventeen or Bust service"
c:
cd "\program files\sb"
sb110

Run the upgrade, press Exit button when the client pops up.

Back to Command Prompt -
net start "Seventeen or Bust service"

*9x/Me*
Command Prompt -
c:
cd "\program files\sb"
sobsvc9x -q
sb110

Run the upgrade, press Exit button when the client pops up.

Back to Command Prompt -
sobsvc9x

BTW: regarding dial-up problems - there is a queueing feature being worked on (for the service handler) that will help with *part* of this. It will eliminate the need to dial-in to get a new test when you've completely finished one.

Unfortunately, my current design doesn't help with the need to transmit intermediate blocks to keep a test alive. I'm headed to Japan for a couple of weeks, so I'll give it a bit of a think and see if I can do anything about that.

[Jammy]

Thanks MathGuy! I hope that your trip to Japan will give you time for a needed rest!

Jammy


Originally posted by MathGuy

Unfortunately, my current design doesn't help with the need to transmit intermediate blocks to keep a test alive. I'm headed to Japan for a couple of weeks, so I'll give it a bit of a think and see if I can do anything about that. [/B]

[MathGuy]

Originally posted by Jammy
Thanks MathGuy! I hope that your trip to Japan will give you time for a needed rest!


Well, I don't know about *rest* - I'm going as a chaperone and tour photographer for a Youth Choir of 37 high-school students (one of whom is my daughter). It'll be lots of fun, though!

I'll chat with all of you folks when I get back...

Assuming the technology works, http://wryc.uwctl.org will have pictures posted each day, so if anyone is interested, you can see probably everyone except me (since I'm tour photographer...)

[Jammy]

Hey . . .I have bookmarked that page and will be looking forward to viewing some pics with your daughter! Maybe you can single her out for us?

Jammy :)




Originally posted by MathGuy


Assuming the technology works, http://wryc.uwctl.org will have pictures posted each day, so if anyone is interested, you can see probably everyone except me (since I'm tour photographer...)

[Marco_N]

The list of fixes in 1.10 has made me confident enough to re-enable intermediate block reporting. So far, so good. I'll keep you guys posted.

- Marco

[Nuri]

Louie,

I guess this version still does not incorporate the enhancements in GIMPS version 23 beta, right?

If so, that's good news. Then there is still room for additional 25% speed increase. I've been running v 23.4.1 for the last ten days, and it seems to work smoothly.

I'd be very happy if you could take a look at if it is possible to incorporate those enhancements to our client.

Thx.

[Nuri]

BTW, I see the "server busy -- block added to submit queue" very frequently with the new client.

Does anybody else have this problem?

[jjjjL]

Originally posted by Nuri
I guess this version still does not incorporate the enhancements in GIMPS version 23 beta, right?


Actually, this version *does* incorporate the enhancements. I know it doesn't seem like it, but it does. I updated all the objects when I made v1.10. I was as surprised as others when I saw that it wasn't any faster until I read into the "improvements" in the GIMPS objs some more. Only the largest FFT sizes were updated. I think the current SB tests are in the 768k FFT right now. GIMPS is higher so I guess GW decided to skip optimizing our size. :(

So the good news is the optimizations are in the client. The bad news is you won't be able to tell until we get to the next (or next after next) FFT crossover.

I don't have a P4 to test speed increases. If you test the new client vs v1 and can show me that it doesn't outperform it for any size exponent, I'll look into it some more. My guess is that if you raised your exponent to about 1/2 the size of current GIMPS tests (n=10M ?) you would get a 25% speed increase.

-Louie

[Keroberts1]

have you tested it with the 3l337 account?

[Marco_N]

Originally posted by Nuri
BTW, I see the "server busy -- block added to submit queue" very frequently with the new client.

Does anybody else have this problem?


I have the same symptoms, however I wouldn't rate it as a problem but rather as a feature, because I suspect this is where the "no record of proth test" (test dropped) problem was before. Louie?

- Marco

[jjjjL]

i think that was where the 99.9% hang problem was.

the code now has recv() timeouts so it decides that it's better to just skip past the block transmission and keep going rather than hang.

the server has been restarted several times lately while I add things. considering that someone is connecting to it every 2 seconds, a few minutes of downtime for upgrades can mean a whole lot of error messages. however, I would agree with Marco that the messages are totally harmless compared to hangs.

-Louie

[Mystwalker]

Is there any projected release date for the new Linux version?
I started the old client (with connect through the proxy server) on 10 machines yesterday. Today, only 2 of them were still running. :cry:

[jjjjL]

Originally posted by Mystwalker
Is there any projected release date for the new Linux version?

no. but i'm thinking about starting it.

[dragongoddess]

according to my log it took 6 hours to do 1 block. What gives?

[Frodo42]

Well just to get Louie from thinking to actually doing it, I also miss the Linux version a lot.
6 computers restart the client once a day and appearently only somewhere around half of them make it sieving to the time where they are stopped (15 hours).