Things we're changed/working on
log in

Advanced search

Message boards : News : Things we're changed/working on

1 · 2 · 3 · Next
Author Message
Profile Kevin
Volunteer moderator
Project administrator
Project developer
Project tester
Volunteer developer
Volunteer tester
Project scientist
Avatar
Send message
Joined: 27 Jul 12
Posts: 507
Credit: 14,550,449
RAC: 2,760
Message 4080 - Posted: 29 Apr 2015, 9:49:45 UTC
Last modified: 29 Apr 2015, 9:50:10 UTC


  • We've implemented the size class so that slower machines (my old XP machine, Androids, etc) get small number of pixels.
  • We're still trying to get the plan class right for Android


____________
Regards
Kevin
-----
International Centre for Radio Astronomy Research

ace_quaker
Send message
Joined: 4 May 14
Posts: 2
Credit: 12,891,271
RAC: 0
Message 4087 - Posted: 30 Apr 2015, 5:32:32 UTC - in response to Message 4080.

How does credit return stack up to previous results/CPU time? Heard reports it was way less.

What else is new in the client, any improvements with more recent processor extensions?

gomeyer
Send message
Joined: 23 Feb 14
Posts: 18
Credit: 100,000,224
RAC: 95,559
Message 4094 - Posted: 30 Apr 2015, 12:33:19 UTC
Last modified: 30 Apr 2015, 12:39:23 UTC

I would give it an additional day or three for numbers to stabilize, but as of now credit looks close. It may be marginally lower but not by much.

I'm comparing daily totals (NOT RAC, that takes weeks to stabilize)

Systems that I'm comparing are running Linux-64, Win 7-64, and Win 8.1-64.

Of course, YMMV.
____________

JH30895
Send message
Joined: 9 Oct 13
Posts: 7
Credit: 73,963,449
RAC: 58,517
Message 4095 - Posted: 30 Apr 2015, 13:45:35 UTC - in response to Message 4094.

I'm not seeing what you are seeing. My machines are running Darwin, and they are only producing about 30% (credit-wise) of what they produced before the hiatus. Results are for fitsedwrapper 4.00. Version 4.01 results are not in yet, but do not appear to be much different.

Profile alexbeckley
Volunteer moderator
Project administrator
Project developer
Project tester
Volunteer developer
Volunteer tester
Project scientist
Send message
Joined: 14 Jan 13
Posts: 345
Credit: 5,696,421
RAC: 0
Message 4099 - Posted: 1 May 2015, 0:15:13 UTC - in response to Message 4095.

Strange I can't see why it would be different between Mac and windows/linux.
Are you looking at RAC values or individual work units and divided credit by cpu seconds?
____________
Cheers
Alex
theSkyNet.org webMaster
ICRAR

JH30895
Send message
Joined: 9 Oct 13
Posts: 7
Credit: 73,963,449
RAC: 58,517
Message 4100 - Posted: 1 May 2015, 3:28:37 UTC - in response to Message 4099.

I no longer have access to work units done prior to the hiatus. What I do have are two dedicated machines that had a daily production of 54,000 (+or- 4,000) and 26,000 (+or- 2,000) credits per day. They are now producing 19,000 and 15,000, respectively (at best). I don't believe that this is unique to Macs, though. My "competition" appears to be constrained as well. I don't see any of the big producers coming anywhere near their former production.

kashi
Send message
Joined: 10 Aug 12
Posts: 44
Credit: 19,419,745
RAC: 6
Message 4101 - Posted: 1 May 2015, 3:50:36 UTC - in response to Message 4100.
Last modified: 1 May 2015, 3:58:02 UTC

Don't know if the number of pending tasks has risen on your i7-3720QM compared to previously but If you're looking at daily credit value on your Xeon CPU for version 4.01 tasks then I'm surprised it hasn't dropped by even more.

Perhaps some kind of homogenous redundancy may have been introduced because 80% or more of the 72 4.01 tasks I have completed so far on my Xeon W3520 remain without the wingman copy sent. Most of these are Unsent Status but some are Inactive Status. I started on Windows but switched to a Mac VM as Darwin is traditionally faster on POGS. Unsent is far too high on both so it is not related to the operating system.

This has been lasting too long, if my pending tasks don't get sent to wingmen within the next day or so I'll switch to another project for a few days and see if things improve. Don't want to be wasting computing resources if it goes kaput.
____________

Profile alexbeckley
Volunteer moderator
Project administrator
Project developer
Project tester
Volunteer developer
Volunteer tester
Project scientist
Send message
Joined: 14 Jan 13
Posts: 345
Credit: 5,696,421
RAC: 0
Message 4103 - Posted: 1 May 2015, 6:41:45 UTC - in response to Message 4101.

We've been having some issues with android clients erroring out and rolling through hundreds of work units which has been stuffing about but I believe i've narrowed down the source of this so things should be improving next week.
____________
Cheers
Alex
theSkyNet.org webMaster
ICRAR

kashi
Send message
Joined: 10 Aug 12
Posts: 44
Credit: 19,419,745
RAC: 6
Message 4104 - Posted: 1 May 2015, 7:42:06 UTC - in response to Message 4103.

Thanks for your reply Alex.

I'm shutting down POGS for now but will leave 11 tasks in cache to "mature". If any of them get a wingman task sent I will unsuspend and process them.

Hope you get it sorted soon.
____________

JH30895
Send message
Joined: 9 Oct 13
Posts: 7
Credit: 73,963,449
RAC: 58,517
Message 4107 - Posted: 1 May 2015, 12:44:01 UTC - in response to Message 4101.

The growing queue in validation pending, that Kashi points out, is a concern, but even if all work units were validated immediately, I calculate that the Xeon would be producing about 33,000 credits per day. That's down from 54,000 before. I strongly suspect that this reduction extends across the board.

TheHoosh
Send message
Joined: 11 Mar 14
Posts: 5
Credit: 10,341,974
RAC: 187
Message 4108 - Posted: 1 May 2015, 17:03:18 UTC

Well, the new v4.01 work units definitely take longer to process on my machines (+50%), but they also result in +50% more credits being granted.
As far as I can tell nothing has changed here.

BigA21
Send message
Joined: 19 Mar 13
Posts: 10
Credit: 101,488
RAC: 0
Message 4111 - Posted: 3 May 2015, 2:25:50 UTC - in response to Message 4108.

Hi everyone, I just wanted to let you know that the 4.0 version works on my Android tablet, it downloaded it fine ran fine and has been validated. However for the Android PIE version and the Android PIE old client version, they downloaded fine ran fine (with a few errors, still trying to figure out why), but haven't been validated yet. As far as I can tell, they should be fine as they have been completed and no noticeable error messages have come up. Does anyone know about how long it will take for my remaining wu's to get validated? As soon as they get validated, I'll post another response stating the outcome.

BitBangerUSA
Send message
Joined: 8 Dec 13
Posts: 12
Credit: 8,603,113
RAC: 0
Message 4116 - Posted: 4 May 2015, 3:37:37 UTC - in response to Message 4111.
Last modified: 4 May 2015, 3:59:01 UTC

WU validation requires that at least one other user process the same data as you. some refer to this other user as a 'wingman.' best you can do is look at your tasks awaiting validation, find the other user(s) and note when they got the task and how much computing horse-power they have for that computer - still, this is not a good indicator as they can be running flat out or only devoting some processor percentage to the task.

BigA21
Send message
Joined: 19 Mar 13
Posts: 10
Credit: 101,488
RAC: 0
Message 4123 - Posted: 5 May 2015, 1:06:09 UTC - in response to Message 4116.

Ok, I just recently checked my count so far, i've got 9 wu's waiting to get validated. This is starting to get out of control, however this is just my opinion. I haven't had this happen in a long time, the most i've seen is 5 to 6 wu's pending. I hope this helps.

Profile Kevin
Volunteer moderator
Project administrator
Project developer
Project tester
Volunteer developer
Volunteer tester
Project scientist
Avatar
Send message
Joined: 27 Jul 12
Posts: 507
Credit: 14,550,449
RAC: 2,760
Message 4124 - Posted: 5 May 2015, 2:08:21 UTC - in response to Message 4123.

I've implemented the size class feature of BOINC. This means that big WUs will go to beefy machines, medium size WUs to mid-range machines and smaller WUs will go to smaller machines. It works off statistics about work unit completion which are calculated each day (so it'll take a few weeks to stablise at where the boundaries are).

This means I have to generate WUs that fit into the 3 categories I have defined: small, medium and large. To start with I did this on a per galaxy basis, but that was a mistake as it generate 10,000s of small WU from a big galaxy. This isn't good as the feeder has to look past all of these to find medium and large WU which is very slow. So I reimplemented it on a per strip per galaxy so a single galaxy can generate all three types by adjusting the pixels per WU and the height of the area.

To get past the 10,000s of little ones I've have had the feeder on random over the weekend to allow the size class WU to flow. Now we're down to a around 250,000 WUs I've switched it back to priority and create time so you should see over the next week the validation sorts itself out.
____________
Regards
Kevin
-----
International Centre for Radio Astronomy Research

BitBangerUSA
Send message
Joined: 8 Dec 13
Posts: 12
Credit: 8,603,113
RAC: 0
Message 4125 - Posted: 5 May 2015, 3:13:58 UTC - in response to Message 4124.

thanks for the info.

the changes certainly invalidate (no pun intended) what i previously posted.

i've noted that i have many WUs waiting for validation that have tasks 'unsent' to other comps. hopefully will be assigned and processed soon.

does sound like you unintentionally created a bottle-neck...

Profile Kevin
Volunteer moderator
Project administrator
Project developer
Project tester
Volunteer developer
Volunteer tester
Project scientist
Avatar
Send message
Joined: 27 Jul 12
Posts: 507
Credit: 14,550,449
RAC: 2,760
Message 4127 - Posted: 5 May 2015, 7:05:43 UTC - in response to Message 4125.

@BitBangerUSA

Yep too many small WUs
____________
Regards
Kevin
-----
International Centre for Radio Astronomy Research

Phil
Avatar
Send message
Joined: 7 Sep 13
Posts: 122
Credit: 2,159,774
RAC: 205
Message 4129 - Posted: 5 May 2015, 10:20:04 UTC - in response to Message 4124.
Last modified: 5 May 2015, 10:32:53 UTC

I've implemented the size class feature of BOINC. This means that big WUs will go to beefy machines, medium size WUs to mid-range machines and smaller WUs will go to smaller machines. It works off statistics about work unit completion which are calculated each day (so it'll take a few weeks to stablise at where the boundaries are).

This means I have to generate WUs that fit into the 3 categories I have defined: small, medium and large. To start with I did this on a per galaxy basis, but that was a mistake as it generate 10,000s of small WU from a big galaxy. This isn't good as the feeder has to look past all of these to find medium and large WU which is very slow. So I reimplemented it on a per strip per galaxy so a single galaxy can generate all three types by adjusting the pixels per WU and the height of the area.


Aah! Thats why I had so many unsent/inactive partnered jobs.
Glad to know It'll All Sort Itself Out.
A lot of BOINC projects grant credits quickly, a kind of "Instant Gratification", and volunteers forget in projects like SETI@Home you might have to wait 2 or 3 weeks for credit for a specific work unit.

Cameron
Send message
Joined: 20 Oct 12
Posts: 19
Credit: 508,089
RAC: 0
Message 4133 - Posted: 6 May 2015, 1:10:40 UTC

How many pixels are in each small, medium and large workunit?

-- Cameron

BigA21
Send message
Joined: 19 Mar 13
Posts: 10
Credit: 101,488
RAC: 0
Message 4139 - Posted: 7 May 2015, 4:08:24 UTC - in response to Message 4133.

Hi everyone, I was just wondering what percentage of users crunch large wu's, medium wu's and small wu's. I can say that there are more people crunching small wu's. Maybe the system that distributes the wu's can be adaptive, by breaking up a typical galaxy image into some large wu's, more medium wu's and many small wu's, as well as taking into account the relative proportions of each. For example let's say the proportion of large users to medium to small as percentages are: 20%, 30%, and 50% (this is just an estimate as I don't know what the actual numbers are). Also a weighting factor can be applied to each class of wu so you can get a weighted average for the typical wu distribution. What are your thoughts?

1 · 2 · 3 · Next

Message boards : News : Things we're changed/working on


Main page · Your account · Message boards


Copyright © 2017 The International Centre for Radio Astronomy Research