Greetings,

I have noticed something weird going on concerning BOINC and programs I use on a daily basis, especially online games and a standalone. This since the addition of the Breakthrough WUs.

IN THE PAST:
Used to be that when I played Star Trek Online I would have to snooze BOINC or shut it down completely due to jerkiness caused in the display. When playing World of Warcraft there was no problem; BOINC could run right alongside WoW without a problem.

I also have a non-Micro$oft solitaire game I play now and then and BOINC had no problem running alongside it.


IN THE NOW:
Now, since the introduction of the Breakthrough WUs, I have to snooze or shut down BOINC completely depending on my time spent in game. Even WoW and the solitaire game have a problem with BOINC running now.

My partner and I run WoW for about 3 to 4 hours in the afternoon (one of the perks of being retired ;) ). I have to shut down BOINC or WoW becomes to jerky for comfort meaning that running on the ground or flying sucks.


Are the Breakthrough WUs causing BOINC to snatch more CPU cycles from other running programs?

Any thoughts?

Keep on BOINCing...! :) (I'm trying... ;) )

---

CAPT Siran d'Vel'nahr - L L & P _\\//
Winders 11 OS? "What a piece of junk!" - L. Skywalker
"Logic is the cement of our civilization with which we ascend from chaos using reason as our guide." - T'Plana-hath

It might be due to the high ratio of GBT data that generates VLARs. I've always noticed on my machines, that were CPU only, when all cores/threads were loaded with VLAR tasks the system would have noticeably lag. On the 12c/24t server I was running I had to limit the number of MB tasks. As 24 VLARs made the server unresponsive for the network users.

If there was a was to assigned a weighted value to VLARs I would set it ti 1.3-1.5 depending on the machine. So that fewer ran at once.

---

SETI@home classic workunits: 93,865 CPU time: 863,447 hours
Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[

It might be due to the high ratio of GBT data that generates VLARs. I've always noticed on my machines, that were CPU only, when all cores/threads were loaded with VLAR tasks the system would have noticeably lag. On the 12c/24t server I was running I had to limit the number of MB tasks. As 24 VLARs made the server unresponsive for the network users.

If there was a was to assigned a weighted value to VLARs I would set it ti 1.3-1.5 depending on the machine. So that fewer ran at once.

That's quite interesting effect of VLAR cause CPU, instead of GPU, has preemptive multitasking. Also, CPU SETI process (again, instead of GPU one) run at idle priority so will should be preemted virtually by everything.
So, if such process still can create noticeable lags it means OS thread scheduling mechanism unable to do its work right for some reason.
And most interesting that this inability depends from process' data (VLAR/non-VLAR).
The single entity that can have such data dependance that comes to mind is memory subsystem. VLAR can have noticeably different cache miss ratio than other datasets (another reason of slowdown on VLAR for modern GPUs that usually have cache but smaller than CPU one).
So, it seems in memory bus saturated scenario OS threading subsystem can't perform its task scheduling correctly (that is, to obey thread priorities).

BTW, what OS your 12c/24t server uses?

It might be due to the high ratio of GBT data that generates VLARs. I've always noticed on my machines, that were CPU only, when all cores/threads were loaded with VLAR tasks the system would have noticeably lag. On the 12c/24t server I was running I had to limit the number of MB tasks. As 24 VLARs made the server unresponsive for the network users.

If there was a was to assigned a weighted value to VLARs I would set it ti 1.3-1.5 depending on the machine. So that fewer ran at once.

That's quite interesting effect of VLAR cause CPU, instead of GPU, has preemptive multitasking. Also, CPU SETI process (again, instead of GPU one) run at idle priority so will should be preemted virtually by everything.
So, if such process still can create noticeable lags it means OS thread scheduling mechanism unable to do its work right for some reason.
And most interesting that this inability depends from process' data (VLAR/non-VLAR).
The single entity that can have such data dependance that comes to mind is memory subsystem. VLAR can have noticeably different cache miss ratio than other datasets (another reason of slowdown on VLAR for modern GPUs that usually have cache but smaller than CPU one).
So, it seems in memory bus saturated scenario OS threading subsystem can't perform its task scheduling correctly (that is, to obey thread priorities).

BTW, what OS your 12c/24t server uses?

That machine was running Windows Server 2008 R2. The CPUs were a pair of Xeon E5645 with 3 4GB DIMMs per processor for 24GB total. I don't recall which app I was running on it, but it would have been AKv8c_r2549_winx86-64_AVXxjfs.exe or AKv8c_r2549_winx86-64_SSSE3xjfs.exe.
I've noticed lag issues on other systems as well when fulled loaded with VLARs. Such as when watching youtube videos in 1080p. Which was one of the reasons I had several of the systems I interacted with set to only run AP work. While systems I didn't use much would run only MB.

I believe someone mentioned at one point VLARs were more memory heavy compared to normal AR tasks. Perhaps that is due to the higher cache miss ratio you mentioned? So systems that already have high memory usage would likely see more noticeable effects?

---

SETI@home classic workunits: 93,865 CPU time: 863,447 hours
Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[

I believe someone mentioned at one point VLARs were more memory heavy compared to normal AR tasks. Perhaps that is due to the higher cache miss ratio you mentioned? So systems that already have high memory usage would likely see more noticeable effects?

Seems so. Strange that AP doesn't cause similar effects though.
PulseFind in MB quite similar to array folding used in AP.
And in VLAR PulseFind's array are longest ones with increased share of them. And the longer folding array the more chances its folding will cause constant cache thrashing at least on initial stages.
From other side, task switching operates on ms scale (quantum in Windows ~20ms) while even cache miss should be on us scale at least.
That is, not quite clear how this VLAR/non-VLAr difference even if exist could be responsive for human-noticeable lags (hundreds of ms scale, I suppose).

EDIT: would be interesting to monitor MB process resource consumption for varius resources (disk IO, memory IO, DPC/interrupts, cache misses and so on) between VLAR/non-VLAR transition. Cause VLAR dataset size is exactly the same 1MB as for non-VLAR task. What is different is how often "non-square" matrixes occur in that data in processing.

It might be due to the high ratio of GBT data that generates VLARs. I've always noticed on my machines, that were CPU only, when all cores/threads were loaded with VLAR tasks the system would have noticeably lag. On the 12c/24t server I was running I had to limit the number of MB tasks. As 24 VLARs made the server unresponsive for the network users.

If there was a was to assigned a weighted value to VLARs I would set it ti 1.3-1.5 depending on the machine. So that fewer ran at once.

Greetings Hal,

Thanks. :)

I just looked at my WU cache a bit closer and a good 90% or more of it is GBT WUs. And they are all VLARs. I just wish we got more AP WUs than we do, I would go all AP then. Right now my PC would be sitting idle to much until more APs were split.

I guess I will just deal with the lag by shutting down BOINC when need be.

Thanks again! :)

Keep on BOINCing...! :)

---

CAPT Siran d'Vel'nahr - L L & P _\\//
Winders 11 OS? "What a piece of junk!" - L. Skywalker
"Logic is the cement of our civilization with which we ascend from chaos using reason as our guide." - T'Plana-hath

It might be due to the high ratio of GBT data that generates VLARs. I've always noticed on my machines, that were CPU only, when all cores/threads were loaded with VLAR tasks the system would have noticeably lag. On the 12c/24t server I was running I had to limit the number of MB tasks. As 24 VLARs made the server unresponsive for the network users.

If there was a was to assigned a weighted value to VLARs I would set it ti 1.3-1.5 depending on the machine. So that fewer ran at once.

Greetings Hal,

Thanks. :)

I just looked at my WU cache a bit closer and a good 90% or more of it is GBT WUs. And they are all VLARs. I just wish we got more AP WUs than we do, I would go all AP then. Right now my PC would be sitting idle to much until more APs were split.

I guess I will just deal with the lag by shutting down BOINC when need be.

Thanks again! :)

Keep on BOINCing...! :)

You might want to look at using <exclusive_app> or <exclusive_gpu_app> in your cc_config.xml

---

SETI@home classic workunits: 93,865 CPU time: 863,447 hours
Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[

It might be due to the high ratio of GBT data that generates VLARs. I've always noticed on my machines, that were CPU only, when all cores/threads were loaded with VLAR tasks the system would have noticeably lag. On the 12c/24t server I was running I had to limit the number of MB tasks. As 24 VLARs made the server unresponsive for the network users.

If there was a was to assigned a weighted value to VLARs I would set it ti 1.3-1.5 depending on the machine. So that fewer ran at once.

Greetings Hal,

Thanks. :)

I just looked at my WU cache a bit closer and a good 90% or more of it is GBT WUs. And they are all VLARs. I just wish we got more AP WUs than we do, I would go all AP then. Right now my PC would be sitting idle to much until more APs were split.

I guess I will just deal with the lag by shutting down BOINC when need be.

Thanks again! :)

Keep on BOINCing...! :)

You might want to look at using <exclusive_app> or <exclusive_gpu_app> in your cc_config.xml

Greetings Hal,

Thanks! I will look into that. :)

Keep on BOINCing...! :)

---

CAPT Siran d'Vel'nahr - L L & P _\\//
Winders 11 OS? "What a piece of junk!" - L. Skywalker
"Logic is the cement of our civilization with which we ascend from chaos using reason as our guide." - T'Plana-hath

Greetings,

@HAL9000:
Took a bit to get it to work, but it works as required. This is what I came up with:

<?xml version="1.0" encoding="ISO-8859-1" ?>
<cc_config>
	<log_flags>
	</log_flags>
	<options>
		<exclusive_app>Wow.exe</exclusive_app>
		<exclusive_app>Wow-64.exe</exclusive_app>
		<exclusive_app>Star Trek Online.exe</exclusive_app>
		<exclusive_app>123FreeSolitaire.exe</exclusive_app>
	</options>
</cc_config>

I googled examples on the Internet to get this figured out. The page you linked to should say that each app needs to be on a separate line in the file.

I wasn't sure if each app was to be delimited on one line so I experimented. First I used " | " between each file. The event log showed the config line with the files names delimited with the vertical bar. That didn't work. So, I tried a ", " delimitation. Again the event log showed the config line, but still no go. So then I did as the code above and after testing I had a "Eureka!" moment. :) It worked.

Yeah I know. On something like this I could have just asked here how to do it. But, sometimes I just like to figure things out myself. Ya know? :)

So now I don't have to snooze BOINC or shut it down and forget to restart it. It is taken care of automagically. :)

Thanks for the idea Hal. :)

Keep on BOINCing...! :)

[edit]Edited the topic.[/edit]

---

CAPT Siran d'Vel'nahr - L L & P _\\//
Winders 11 OS? "What a piece of junk!" - L. Skywalker
"Logic is the cement of our civilization with which we ascend from chaos using reason as our guide." - T'Plana-hath

The page you linked to should say that each app needs to be on a separate line in the file.

Done.

The page you linked to should say that each app needs to be on a separate line in the file.

Done.

Greetings Richard,

Wow! Cool! Thanks! :) :) :)

Keep on BOINCing...! :)

---

CAPT Siran d'Vel'nahr - L L & P _\\//
Winders 11 OS? "What a piece of junk!" - L. Skywalker
"Logic is the cement of our civilization with which we ascend from chaos using reason as our guide." - T'Plana-hath

I place multiple elements on a single line in mine without any problems. I think the issue you had was using non-XML separators between elements..

<options>
	<exclusive_app>TS3W.exe</exclusive_app><exclusive_app>fallout4.exe</exclusive_app>
	<exclusive_gpu_app>TS3W.exe</exclusive_gpu_app><exclusive_gpu_app>fallout4.exe</exclusive_gpu_app>
</options>

09-May-2016 17:52:32 [---] Config: don't compute while TS3W.exe is running
09-May-2016 17:52:32 [---] Config: don't compute while fallout4.exe is running
09-May-2016 17:52:32 [---] Config: don't use GPUs while TS3W.exe is running
09-May-2016 17:52:32 [---] Config: don't use GPUs while fallout4.exe is running

Note: My use of <exclusive_app> & <exclusive_gpu_app> is redundant. However I was originally using <exclusive_gpu_app> then found I needed to have CPU apps suspended as well for fallout. So I copied the <exclusive_gpu_app> line to <exclusive_app>. Thinking it would be easier to comment it out if I stopped running CPU tasks. For reference XML comments are done with start and end tags line this <!--something goes here--> & can span multiple lines.
Also I imagine I could have placed all 4 of my exclusions on one line, but I prefer to have CPU & GPU exclusions on separate lines.

---

SETI@home classic workunits: 93,865 CPU time: 863,447 hours
Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[

I place multiple elements on a single line in mine without any problems. I think the issue you had was using non-XML separators between elements..

<options>
	<exclusive_app>TS3W.exe</exclusive_app><exclusive_app>fallout4.exe</exclusive_app>
	<exclusive_gpu_app>TS3W.exe</exclusive_gpu_app><exclusive_gpu_app>fallout4.exe</exclusive_gpu_app>
</options>

09-May-2016 17:52:32 [---] Config: don't compute while TS3W.exe is running
09-May-2016 17:52:32 [---] Config: don't compute while fallout4.exe is running
09-May-2016 17:52:32 [---] Config: don't use GPUs while TS3W.exe is running
09-May-2016 17:52:32 [---] Config: don't use GPUs while fallout4.exe is running

Note: My use of <exclusive_app> & <exclusive_gpu_app> is redundant. However I was originally using <exclusive_gpu_app> then found I needed to have CPU apps suspended as well for fallout. So I copied the <exclusive_gpu_app> line to <exclusive_app>. Thinking it would be easier to comment it out if I stopped running CPU tasks. For reference XML comments are done with start and end tags line this <!--something goes here--> & can span multiple lines.
Also I imagine I could have placed all 4 of my exclusions on one line, but I prefer to have CPU & GPU exclusions on separate lines.

Greetings Hal,

This was how I was trying to delimit the filenames:

<exclusive_app>filename1.exe | filename2.exe | ...</exclusive_app>

and the commas:

<exclusive_app>filename1.exe, filename2.exe, ...</exclusive_app>

They showed up in the event log as:

Config: don't compute while filename1.exe | filename2.exe | ... is running
Config: don't compute while filename1.exe, filename2.exe, ... is running

The comment line code is the same as in HTML. :)

Thing is, I saw an example of GPUs delimited with the vertical bar which gave me the idea. Which of course, in this case, was wrong. ;)

Anyway, it's working just fine now. Thanks again! :)

Keep on BOINCing...! :)

---

CAPT Siran d'Vel'nahr - L L & P _\\//
Winders 11 OS? "What a piece of junk!" - L. Skywalker
"Logic is the cement of our civilization with which we ascend from chaos using reason as our guide." - T'Plana-hath

I place multiple elements on a single line in mine without any problems. I think the issue you had was using non-XML separators between elements..

<options>
	<exclusive_app>TS3W.exe</exclusive_app><exclusive_app>fallout4.exe</exclusive_app>
	<exclusive_gpu_app>TS3W.exe</exclusive_gpu_app><exclusive_gpu_app>fallout4.exe</exclusive_gpu_app>
</options>

09-May-2016 17:52:32 [---] Config: don't compute while TS3W.exe is running
09-May-2016 17:52:32 [---] Config: don't compute while fallout4.exe is running
09-May-2016 17:52:32 [---] Config: don't use GPUs while TS3W.exe is running
09-May-2016 17:52:32 [---] Config: don't use GPUs while fallout4.exe is running

Note: My use of <exclusive_app> & <exclusive_gpu_app> is redundant. However I was originally using <exclusive_gpu_app> then found I needed to have CPU apps suspended as well for fallout. So I copied the <exclusive_gpu_app> line to <exclusive_app>. Thinking it would be easier to comment it out if I stopped running CPU tasks. For reference XML comments are done with start and end tags line this <!--something goes here--> & can span multiple lines.
Also I imagine I could have placed all 4 of my exclusions on one line, but I prefer to have CPU & GPU exclusions on separate lines.

Greetings Hal,

This was how I was trying to delimit the filenames:

<exclusive_app>filename1.exe | filename2.exe | ...</exclusive_app>

and the commas:

<exclusive_app>filename1.exe, filename2.exe, ...</exclusive_app>

They showed up in the event log as:

Config: don't compute while filename1.exe | filename2.exe | ... is running
Config: don't compute while filename1.exe, filename2.exe, ... is running

The comment line code is the same as in HTML. :)

Thing is, I saw an example of GPUs delimited with the vertical bar which gave me the idea. Which of course, in this case, was wrong. ;)

Anyway, it's working just fine now. Thanks again! :)

Keep on BOINCing...! :)

Ah I see what you were doing. I think maybe it would be most correct to say each application needs to be enclosed in its own set of tags. It is probably possible to make a cc_config.xml that is one single line. That would be hard for humans to read in an app like notepad, but would be displayed correctly in an XML viewer.

I would guess when you saw an example of things separated by a vertical bar, or pipe*, it was something like this.
*I tend to call it a pipe even when I'm not piping a command.

<exclude_gpu>
   <url>project_URL</url>
   [<device_num>N</device_num>]
   [<type>NVIDIA|ATI|intel_gpu</type>]
   [<app>appname</app>]
</exclude_gpu>

Which in the example is meant to display the available options & I guess could be read an "or". <type>NVIDIA or ATI or intel_gpu</type>

It looks like the exclusive app function reads every character between the tags. Which would hopefully work for applications that have a space in their name such as "my app.exe".

---

SETI@home classic workunits: 93,865 CPU time: 863,447 hours
Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[