Beta testing - Calibrated Boinc core
Moderátoři: zdespi, Moderátoři
Beta testing - Calibrated Boinc core
Currently I do not have the possibility to open a non-public thread here, so in the time being, before one of the admins can do it, please send me a private message (see the button SZ or PM under this post) if you are interested in beta testing the Boinc core client
Naposledy upravil(a) trux dne pát led 13, 2006 6:46 pm, celkem upraveno 1 x.
For those who already downloaded the client: please do not post the URL here, nor anywhere else. Please, do not share the software yet. I need to have the installations under control, for case of serious bugs and the necessity to recall the client. I'll probably release it publicly within few days (if no glitches appear, and if the new features work reliably).
Please note that you have to enable the calibration in the configuration, otherwise nothing changes with the claimed credit. Also, it may take up to several days, before the claimed credit reaches the desired level.
I did not really test if the calibration works fine with other projects than S@H. However, you shpould be aware that it is expecte that with calibration enabled at other projects than S@H, it will likely claim lower credit than uncalibrated.
PS: you are welcome to report your concerns, or satisfaction, or report bugs, both here and on the official S@H forum in this thread. What currently needs more testing, is especially the credit calibrating for other projects than S@H. Of course, you should be aware that some projects (i.e. CPDN) use different credit systems and completely ignore the claimed credit.
Please note that you have to enable the calibration in the configuration, otherwise nothing changes with the claimed credit. Also, it may take up to several days, before the claimed credit reaches the desired level.
I did not really test if the calibration works fine with other projects than S@H. However, you shpould be aware that it is expecte that with calibration enabled at other projects than S@H, it will likely claim lower credit than uncalibrated.
PS: you are welcome to report your concerns, or satisfaction, or report bugs, both here and on the official S@H forum in this thread. What currently needs more testing, is especially the credit calibrating for other projects than S@H. Of course, you should be aware that some projects (i.e. CPDN) use different credit systems and completely ignore the claimed credit.
machine: Dual Xeon, w2k adv.server
BOINCmanager: 5.2.13
BOINC core: 5.3.8 tx23
when I place the additional options in the global_preferences block (as in your example), the options are not read on startup.
after updating preferences from a project the boinc client injects an empty truXoft_conf section.
when putting the additional options in the active venue section ("home" for this machine) the options are read and used.
after updating the preferences from the project again, the options are injected into the global_preferences block by the boinc client and not loaded after a restart of the client.
BOINCmanager: 5.2.13
BOINC core: 5.3.8 tx23
when I place the additional options in the global_preferences block (as in your example), the options are not read on startup.
after updating preferences from a project the boinc client injects an empty truXoft_conf section.
when putting the additional options in the active venue section ("home" for this machine) the options are read and used.
after updating the preferences from the project again, the options are injected into the global_preferences block by the boinc client and not loaded after a restart of the client.
Kód: Vybrat vše
<venue name="home">
<run_on_batteries/>
<run_if_user_active/>
<idle_time_to_run>3</idle_time_to_run>
<leave_apps_in_memory/>
<cpu_scheduling_period_minutes>60</cpu_scheduling_period_minutes>
<work_buf_min_days>0.05</work_buf_min_days>
<max_cpus>4</max_cpus>
<disk_interval>300</disk_interval>
<disk_max_used_gb>100</disk_max_used_gb>
<disk_max_used_pct>75</disk_max_used_pct>
<disk_min_free_gb>1</disk_min_free_gb>
<vm_max_used_pct>50</vm_max_used_pct>
<max_bytes_sec_down>0</max_bytes_sec_down>
<max_bytes_sec_up>0</max_bytes_sec_up>
<dont_verify_images/>
<truXoft_conf>
<return_results_immediately/>
<cpu_affinity/>
<calibrate_credit>all</calibrate_credit>
</truXoft_conf>
</venue>Installed it on 3 machines now, the XP2400+, the dual MP2400+ and the dual Xeon 2.4.
All three machines have the settings I posted earlier
I picked SETI and Predictor to start looking at the credit calibration.
Right now the claimed credit matches the inflated benchmark results for the machines.
I'll check the results again tomorrow over my morning coffee to see if there is any change in calibration after a number of processed results for both projects.
All three machines have the settings I posted earlier
I picked SETI and Predictor to start looking at the credit calibration.
Right now the claimed credit matches the inflated benchmark results for the machines.
I'll check the results again tomorrow over my morning coffee to see if there is any change in calibration after a number of processed results for both projects.
OK, I am looking forward, although I have no idea what crediting system Predictor uses, and if the calibrating has any chance to influence it.
As for the venues - I looked at it and unfortunately it gets more complicated than I expected. The global parameters, including all the venue settings, are being transfered from the active project server in one huge block that is then used to overwrite the local global_prefs file without parsing it. Currently, I injected the add-on parameters at the beginning, but cannot do the same without first parsing all the block, extracting and processing all parameters and venues, then injecting the add on parameters to the right place, and writing to the file. Although possible, it is a lot of coding and might need to be rewritten at each new synchronizing with the official source code. Therefore, I decided to pull out of the global_prefs.xml file and using a separate configuration file just for my add-ons. I'll keep the backwars compatibility though. Parameters in global_prefs.xml will be accepted and written to the new file anyway.
This is a very unlucky situation, because originally, in my very early clients, I already had all parameters rather conveniently in a simple format in the remote_hosts.cfg file, working without any glitches, and just decided to do it "properly" after getting some complaints about the illogical location of the parameters. In fact, I spent more time in messing up with the configuration, than with the actual modifications. Arrgh! I should stop listening complaints, and better do the things the simple direct way.
As for the venues - I looked at it and unfortunately it gets more complicated than I expected. The global parameters, including all the venue settings, are being transfered from the active project server in one huge block that is then used to overwrite the local global_prefs file without parsing it. Currently, I injected the add-on parameters at the beginning, but cannot do the same without first parsing all the block, extracting and processing all parameters and venues, then injecting the add on parameters to the right place, and writing to the file. Although possible, it is a lot of coding and might need to be rewritten at each new synchronizing with the official source code. Therefore, I decided to pull out of the global_prefs.xml file and using a separate configuration file just for my add-ons. I'll keep the backwars compatibility though. Parameters in global_prefs.xml will be accepted and written to the new file anyway.
This is a very unlucky situation, because originally, in my very early clients, I already had all parameters rather conveniently in a simple format in the remote_hosts.cfg file, working without any glitches, and just decided to do it "properly" after getting some complaints about the illogical location of the parameters. In fact, I spent more time in messing up with the configuration, than with the actual modifications. Arrgh! I should stop listening complaints, and better do the things the simple direct way.
I just uploaded a new release, with the fix of the configuration issue. I finally gave up on trying to make it working correctly with global_prefs.xml. It was far too much of hassle. The file and all routines around it were simply not designed for reading, parsing and storing the settings locally, but just for reading data from the project server in a big block and storing it on the disk. Messing with the venues turned to be too much of a pain, and in fact I was trying to do something what was not really clean anyway - the purpose of the file is simply different. I moved the parameters to a new file that is called truxoft_prefs.xml. If you already have your settings in the global_prefs.xml, you do not need to do anything - after loading the new client it will create the new file and move all add-on settings there. It will aslo accept the very original settings in remote_hosts.cfg in the old format too, so the backwards compatibility is maintained.
I tested it just briefly and it is late here (4:15AM), so please bear with me if you find a new bug.
I tested it just briefly and it is late here (4:15AM), so please bear with me if you find a new bug.
the few results I returned at Predictor with tx23 seem to be claiming according to benchmark.
I installed tx24 just now and removed the additional settings from the global_prefs.xml.
my settings on the dual cpu systems in the truxoft_prefs.xml:
the XP2400+ is running with <calibrate_credit>all</calibrate_credit>
(projectnames copied from the stderrdae.txt file)
[EDIT]
first SETI result uploaded from the MP2400+
first Predictor result from the Dual Xeon
[/EDIT]
I installed tx24 just now and removed the additional settings from the global_prefs.xml.
my settings on the dual cpu systems in the truxoft_prefs.xml:
Kód: Vybrat vše
<truXoft_conf>
<return_results_immediately/>
<cpu_affinity/>
<calibrate_credit>SETI@home,rosetta@home,Predictor @ Home</calibrate_credit>
</truXoft_conf>(projectnames copied from the stderrdae.txt file)
[EDIT]
first SETI result uploaded from the MP2400+
Kód: Vybrat vše
Project Date Message
SETI@home 1/14/2006 10:49:59 AM Computation for result 09fe05aa.29400.16498.117304.1.50_1 finished
seti@home 1/14/2006 10:49:59 AM CC calibration: blocked [negative calibration limit] 38.62 >> 19.94 (time: 8879s >> 4584s / Gfpops: 1.94 >> 1.94)
seti@home 1/14/2006 10:50:01 AM Started upload of 09fe05aa.29400.16498.117304.1.50_1_0
Kód: Vybrat vše
200512861 47894446 14 Jan 2006 3:40:14 UTC 14 Jan 2006 9:47:45 UTC Over Success Done 8,879.22 40.72 pendingfirst Predictor result from the Dual Xeon
Kód: Vybrat vše
1/14/2006 11:17:22 AM|Predictor @ Home|Computation for result h0014B_1_72335_0 finished
1/14/2006 11:17:22 AM|predictor @ home|CC calibration: blocked [negative calibration limit] 11.83 >> 11.27 (time: 6639s >> 6106s / Gfpops: 1.36 >> 1.47)
1/14/2006 11:17:25 AM|predictor @ home|Started upload of h0014B_1_72335_0_0
1/14/2006 11:17:25 AM|predictor @ home|Started upload of h0014B_1_72335_0_1Kód: Vybrat vše
15648625 3753478 14 Jan 2006 6:54:50 UTC 14 Jan 2006 10:18:15 UTC Over Success Done 6,638.72 12.28 pendingfirst predictor result for the machine running with <calibrate_credit>all</calibrate_credit>
claimed credit according to BoincView is 12.84 (in line with the benchmarks)
I have a nasty suspicion the "all" switch isn't working.
[EDIT]
I think the negative calibration limit should be adjusted for Predictor.
Currently I am overclaiming by up to 35% over the granted credit on the Xeon and up to 100% on the MP2400+ when compared to older non-optimzed clients.
[/EDIT]
[EDIT2]
I'll build up a textfile with the calibration-history for the hosts and upload it as soon as it seems to stabilize
[/EDIT2]
[EDIT3]
first SETI result for the machine running with <calibrate_credit>all</calibrate_credit>
"all" is not working for the calibration.
[/EDIT3]
Kód: Vybrat vše
14/01/2006 11:34:34||truXoft add-on: calibrate_credit = all
..
14/01/2006 12:05:01|Predictor @ Home|Computation for result h0014B_1_72980_3 finished
14/01/2006 12:05:03|predictor @ home|Started upload of h0014B_1_72980_3_0
14/01/2006 12:05:03|predictor @ home|Started upload of h0014B_1_72980_3_1Kód: Vybrat vše
15651230 3754123 14 Jan 2006 7:49:40 UTC 14 Jan 2006 11:17:47 UTC Over Success Done 2,861.63 12.84 pending[EDIT]
I think the negative calibration limit should be adjusted for Predictor.
Currently I am overclaiming by up to 35% over the granted credit on the Xeon and up to 100% on the MP2400+ when compared to older non-optimzed clients.
[/EDIT]
[EDIT2]
I'll build up a textfile with the calibration-history for the hosts and upload it as soon as it seems to stabilize
[/EDIT2]
[EDIT3]
first SETI result for the machine running with <calibrate_credit>all</calibrate_credit>
Kód: Vybrat vše
14/01/2006 14:03:23|SETI@home|Computation for result 14dc03aa.10580.22816.222162.1.85_1 finished
14/01/2006 14:03:25|seti@home|Started upload of 14dc03aa.10580.22816.222162.1.85_1_0[/EDIT3]
Naposledy upravil(a) Mr.Pernod dne sob led 14, 2006 3:13 pm, celkem upraveno 1 x.
not enough SETI results on the Athlons yet to say for sure, but the calibration is slowly approaching 30 credits/result.
and even better news:
calibration for SETI is working on the Xeon, the 3rd and 4th returned result are claiming around 30 credits now and it seems to be stabilizing.
I'm going out now to buy me a copy of Call of Duty 2, I'll check for tx25 when I get back.
In the mean time: get enough sleep man, you know you need it
and even better news:
calibration for SETI is working on the Xeon, the 3rd and 4th returned result are claiming around 30 credits now and it seems to be stabilizing.
I'm going out now to buy me a copy of Call of Duty 2, I'll check for tx25 when I get back.
In the mean time: get enough sleep man, you know you need it