Ned and company,

I can't recall how to spell his name, Mr. "Tee" I guess I will have to call him. Well, 18 months of failures may be behind me ... I managed, I think, to get an optimized Science Application to work for an hour without barfing ... still have not finished one work unit yet, but the signs are good.

I did my slowest machine so, I don't know when it will finish ... but before my system died before I even got started. So, thank you for the resources...

I will use this to accumulate *MY* experience as I switch my machines over ... I did this one when it was "dry" so, I made sure that I would not lose work ...

I did capture screen shots. So, when I am further down the road I will start to put up a process like what I have done for the installations of the "Basic" BOINC install ...

I have not figured out (yet) how to display the data so I can get an average processing time to compare as time goes on ... so, I still have that to do ...

---

I followed the instructions from the BOINC Wiki last weekend to install the optimized Windows BOINC client and Seti app from Tetsuji. I use Ned's client and app for Linux. I am impressed with the difference it makes. Thanks to Ned and Tetsuji (sp?) for the optimized clients and to Paul for BOINC Wiki.

Regards,

Pam

Pam,

You are not mad at me for not having pictures yet? :)

This is about the 30th time I tried and it looks like maybe 30th time is the charm ... :)

Maybe it was because I was doing the documenting and hoping to have a major foul up so I could get screen shots of THAT .. Mr. Murphy strikes!

---

Thanks to Ned and Tetsuji (sp?) for the optimized clients and to Paul for BOINC Wiki.

I'll second that. I dropped the P3 Linux client in early this morning and much to my surprise the first WU it grabbed is done, and the next well on it's way. This old slow boat Celeron 1GHZ needs all the help it can. Based on the finished unit, and the progress of the one in crunching, it looks like it's gained in the neighborhood of 2 hours per unit.

Way cool ... I want to get one or two under my belt before I do the next systems. I will likely try one that has work in progress to see if I can bollix that up ...

Oh, and I got a query figured out ... MAN IS IT UGLY ...

Pretty fast though ... of course, I am not selecting any of the optimized records as there ARE NONE YET! :)

But, the query seems to be ok ...

Ah, well ... another bit of froth on my life's wake

---

Glad you've finally decided to join in the fun Paul :)

I remember back when I first started making my optimized clients... It was a lonely road, just me stumbling around in the dark. But there were a few bright sparks (WerZ and Benher spring to mind) who helped light my path.

But it wasn't a lonely road for long. Peter Smithson dropped in and did some great work fixing the boinc benchmark over optimization bug. Then Chris Bossard totally renewed my flaining interest and got us back on track on the AMD side. Metod picked up the slack and made great progress with clients for intel processors and of course "Maverick" stole the day with his wonderful Windows clients.

Now we have a whole community helping optimize boinc and seti. I've also had help from a small army of beta testers who have benchmarked all my optimized clients along the way to help identify the best (there have been many many more versions that were'nt ever released). I just hope everyone else has got as much enjoyment out of it as I have. I guess this is my way of saying thank you to all those who have helped along the way :)

Unfortunately though, the time may be getting near when I must step back and let others carry it forward. When I started I was unemployed and had plenty of time to dedicate to it. Now I just don't have the time anymore - I would often spend 20-30 hours/week on it but I need to earn a living to support my family too. I can't see that I can really improve on the optimized clients I've now got and I know I simply don't have the time to dedicate to doing decent versions for future releases of boinc and seti.

My only hope is that those who carry it forward do so in the spirit of boinc - keep it totally open, share it and document it so others can follow :)

Happy crunching,

Ned

---

*** My Guide to Compiling Optimised BOINC and SETI Clients ***
*** Download Optimised BOINC and SETI Clients for Linux Here ***

Well, hand optimization is fine, but making the 5-6 different versions should not be too bad. I wanted to jump in, but, I have my own troubles that take most of my time ... :)

I have to admit that each time I do a major rebuild the quality goes up ... we did not have a good set of processes to help people do things. We still don't have all the ones I want. But, I am making some inroads on the lack ... I think we have about 20 and I have material for about another 10 in the wings ... I just have to get the nose back to the grindstone ...

Oh, finished the first work unit ... now to see if it validates (and no, I did not do the test WU best test is a live fire) ... will take the rest of the week before we will have a decent set of data to work with though (assuming that we vaildate) ... but the preliminary times look good ...

---

I think the work the folks have done in optimising the seti app and the boinc client is absolutely amazing to say the least. I can keep my seti resource share as it was and more than double it's RAC. Now if I was a real credit whore I'd suspend all other projects and only crunch seti, but then that would be no fun. You folks rock!

Just a question....I tend to run at the bleeding edge of development releases (I stopped doing that for about a week when the optimised seti app came out) and so I can not run the optimised boinc client anymore. Once the BOINC V4.7x series of releases hits the main stream will the optimisation guru's release an optimised BOINC V4.7x client?

Live long and crunch.

---

Paul
(S@H1 8888)
And proud of it!

I think the work the folks have done in optimising the seti app and the boinc client is absolutely amazing to say the least. I can keep my seti resource share as it was and more than double it's RAC. Now if I was a real credit whore I'd suspend all other projects and only crunch seti, but then that would be no fun. You folks rock!

Just a question....I tend to run at the bleeding edge of development releases (I stopped doing that for about a week when the optimised seti app came out) and so I can not run the optimised boinc client anymore. Once the BOINC V4.7x series of releases hits the main stream will the optimisation guru's release an optimised BOINC V4.7x client?

Live long and crunch.

I would like to second the request for an optimised Boinc client for 4.72 if it is going to be around long enough to be worth while. Right now I am using the 4.72 manager with the optimised 4.45 client. I would like to get that modifying bench mark so that I can get more than 3 days cache with a setting of 10 days. With the stock setup and a 1.5 P4, I was bench marking at 9 hours and finishing in 6. With the optimised setup (Boinc and Seti), I am bench marking at 7.5 and averaging 2.5 per WU. My RAC has doubled with this computer.

---

If you are REALLY interested in crdit, you wouldn't run SETI@Home anyway. I paused most of my machines to run down the buffers both because of the hiccup and becasue I wanted to grab pictures of the install process. So, what do I see? My actual values rising faster when I am doing less SETI@Home. So, my world standing is doing a lot better while my SETI@Home standing is falling.

Based on some of the other discussions, this is one of those areas where in a month we should be able to make some objective measures ... I guess I should also make a guide to collecting your own stats so others can do the same thing I do with the BOINC View logs and MySQL ...

Right now my database for making analysis is getting nice and large, almost 22,000 work units for the 5 production projects. :)

Doing the tests on requested has to go on the other database and I am not sure that my stuff there is good enough quality for me to release that ... the other one though, that is pretty easy. But all of that is for the future ... I have to do some processing to get there ...

---

If you are REALLY interested in crdit, you wouldn't run SETI@Home anyway. I paused most of my machines to run down the buffers both because of the hiccup and becasue I wanted to grab pictures of the install process. So, what do I see? My actual values rising faster when I am doing less SETI@Home. So, my world standing is doing a lot better while my SETI@Home standing is falling.

...

Geez, I would have thought that if you were 'only' interested in credit then you would soley run the optimised 4.45 client and the suitable seti app. The seti app gets the wu completed in 50% of the time the un-optimised app does and the optimised client gets a little more credit. Let's do a little math (for my machine).

1. Before optimisation. Granted credit = 20 to 30 per wu and a wu takes say 4hrs. So let's average that out to 150 credits a day.

2. Post optimisation. Granted credit = 17 to 30 and a wu takes 2hrs. So this averages out (using an average granted of say 23cs) to 270 credits a day.

In this case 2 is greater than 1 for my machine.

If I run 100% LHC on the same machine it takes 5hrs to do a wu and I get an average of 32cs per wu, so this averages out to around 150cs a day.

All up I'd say if you were purely interested in credits then using the optimised seti app gives the best bang for your cpu cycle hands down.

Live long and crunch.

Paul.

I would like to second the request for an optimised Boinc client for 4.72 if it is going to be around long enough to be worth while. Right now I am using the 4.72 manager with the optimised 4.45 client. I would like to get that modifying bench mark so that I can get more than 3 days cache with a setting of 10 days. With the stock setup and a 1.5 P4, I was bench marking at 9 hours and finishing in 6. With the optimised setup (Boinc and Seti), I am bench marking at 7.5 and averaging 2.5 per WU. My RAC has doubled with this computer.

The latest versions of BOINC clients (4.7x) have a mechanism to adjust the predicted time to run a workunit. On my P-M 2GHz it used to be something like 5 hours while they got crunched (using Tetsuji's optimized science app) in 1:15. After running 4.71 (now 4.72) for a week or so the predicted run time wor a WU dropped to 2:30 which means I have a slightly more accurate amount of work cached.

BTW, I tried to compile the 4.72 CC on Linux using Intel CC ... and failed. I don't have too much spare time so I didn't look into it (yet).

---

Metod ...

Now, if I could get you to do that for OS-X and Windows ... :)

I suspect that I will be, probably, going to be running the optimized SETI@Home Science Application as optimized, but the BOINC Daemon as the later generation. Oddly enough I like the 4.4x versions better. If I could summon my courage I might even be tempted to try the current 4.72 ... But, more likely I will wait for the next "official" recommended release ...

Just for those on the edge of their seats on this, I have completed two work units on the "test" machine and have them reported, still in "pending" state though. Each one also has one other result in ...

I also put that one on the wireless network too ... so I will be moving it to another room later today.

---

1. Before optimisation. Granted credit = 20 to 30 per wu and a wu takes say 4hrs. So let's average that out to 150 credits a day.

2. Post optimisation. Granted credit = 17 to 30 and a wu takes 2hrs. So this averages out (using an average granted of say 23cs) to 270 credits a day.

In this case 2 is greater than 1 for my machine.

If I run 100% LHC on the same machine it takes 5hrs to do a wu and I get an average of 32cs per wu, so this averages out to around 150cs a day.

All up I'd say if you were purely interested in credits then using the optimised seti app gives the best bang for your cpu cycle hands down.

Live long and crunch.

If you've got a decent fast computer I'll show you another math:
CPDN does give you 94.5175 credits every trickle.
1 trickle is done in 10802 timesteps.
My computer (AMD 64 3000+) does a timesteps in ca. 2.31 seconds
...a Day has 86400 seconds...

86400 / 10802 * 94.5175 / 2.31 = 327,27 credits/day

ok, with my optimized client seti@home WUs least about 1:25 (about 1.4h) that makes 24/1.4*23=394credits/day...

better than CPDN, but not that much.

Andy

---

Want to know your pending credit?


The biggest bug is sitting 10 inch in front of the screen.

Yes, for the moment I see the opposite with the non-optimized client. As I only have one system running with the optimized client it will take me a little bit of time to nail this one down. But, preliminary signs look good.

I now have 3 Work Units processed with the optimized client, but none have been validatated yet. so I still have that river to cross ...

---

Paul, you don't need to worry about them validating because that part has been well tested :)

---

98SE XP2500+ @ 2.1 GHz Boinc v5.8.8

And God said"Let there be light."But then the program crashed because he was trying to access the 'light' property of a NULL universe pointer.

Un huh ...

I have heard that before ... just before a THUNK!

But, you are right. I have 4 or 5 back and they all validated. Of course, in the spirit of debugging the install ... I blew up a couple strings of work ... but, I had drained all but two Computers. So, for those that get upset with accidental loss of work units. Feel free to yell ... I don't pay that much attention to people that yell ...

But, all part of the grand expiriment.

Even better, my ISP was doing maintenance last night and I was up all night, so, I finished up some hardware changes (moved sound system to my new work windows work machine, put RAIDSERVER on wireless lan (also reinstalled Windows as the SCSI drive was acting odd, so put in spare IDE) also move all but two computers out of the computer room to see if it will stay comfortable in the afternoons.

Though I am still having a little problem with the RAID SERVER computer coming up on the network. It can see the internet, but none of my computers can see him ... odd ...

{edit}

pinged back from there and we are happier now ...

---

I couldn't decide where to stick it, so I chose this thread.

I've compiled BOINC CC version 4.72 for Linux using ICC. I produced only two different binaries: one optimized for P4 with SSE2 and one generic. Benchmarks on my boxen showed that generic binary is only slightly slower than processor speciffic binary on anything but P4.

It's working fine for me now for 2 days, UL, DL and reporting is fine for all projects I'm running (see my sig).

How to use it?
First download and install the official BOINC suite for Linux. I guess you should get version 4.72 (if you don't know how to get it, you're better off not installing my optimized version anyway). Then download appropriate binary form my site. gunzip it, palce it into the right directory and rename it to boinc (before that, save the official binary just to be safe).

---

Metod ...

I couldn't decide where to stick it, so I chose this thread.

I've compiled BOINC CC version 4.72 for Linux using ICC. I produced only two different binaries: one optimized for P4 with SSE2 and one generic. Benchmarks on my boxen showed that generic binary is only slightly slower than processor speciffic binary on anything but P4.

It's working fine for me now for 2 days, UL, DL and reporting is fine for all projects I'm running (see my sig).

I found the same thing Metod - processor specific optimizations do very little for the boinc core client and my i686 client is nearly as fast as any processor-specific client.

That said, I've also just posted a 32-bit client (based on v4.19) for AMD64 processors for those users running a 32-bit version of linux (x86) on their 64-bit processors.

One quick related question - with my v4.19 boinc clients, if I lose my internet connection for any reason, SETI keeps exiting and restarting the current unit (see below). Just wondered if anyone had seen anything similar or knows the fix - maybe it's fixed in a later revision?



06:16:54 [---] Can't resolve hostname setiboincdata.ssl.berkeley.edu (host not found or server failure)
06:16:54 [---] Can't resolve hostname setiboincdata.ssl.berkeley.edu (host not found or server failure)
06:16:54 [SETI@home] Couldn't start upload for http://setiboincdata.ssl.berkeley.edu/sah_cgi/file_upload_handler: error -113
06:16:54 [SETI@home] Couldn't start upload for http://setiboincdata.ssl.berkeley.edu/sah_cgi/file_upload_handler: error -113
06:16:54 [SETI@home] Backing off 3 minutes and 22 seconds on transfer of file 17ja05aa.11111.30849.940902.20_0_0
06:16:55 [SETI@home] Result 08fe05ac.20599.11568.847134.137_0 exited with zero status but no 'finished' file
06:16:55 [SETI@home] If this happens repeatedly you may need to reset the project.
06:16:55 [SETI@home] Restarting result 08fe05ac.20599.11568.847134.137_0 using setiathome version 4.70

The above sequence happens repeatedly every 5 mins or so until the internet connection is restored.

Ned

---

*** My Guide to Compiling Optimised BOINC and SETI Clients ***
*** Download Optimised BOINC and SETI Clients for Linux Here ***

One quick related question - with my v4.19 boinc clients, if I lose my internet connection for any reason, SETI keeps exiting and restarting the current unit (see below). Just wondered if anyone had seen anything similar or knows the fix - maybe it's fixed in a later revision?

That could be coincidence. I get these (process exited but no finish file) from time to time (and have been running 4.44 for a while until I recently upgraded my crunchers to 4.7x).

Last time I noticed a similar thing with 4.71 on Windows box: the process exited but CC didn't notice that :(. The fix was to restart BOINC CC (but I still lost a night worth of CPU time).

---

Metod ...