BRP4 1.33/1.34 app release: feedback thread

Bernd Machenschalk
Bernd Machenschalk
Moderator
Administrator
Joined: 15 Oct 04
Posts: 4267
Credit: 244932143
RAC: 16370

Compression works, but by

Compression works, but by nature depends on the data. It looks like the apparently quieter "outer galaxy" beams compress much better than the "inner galaxy" ones we got recently.

I'm tracking the compression ratio of some older beams, I've seen everything from 80% down to 25%.

The network output of our download servers dropped to about a third (33%), which looks like it could well be the average of the compression ratios I've seen so far.

BM

BM

Gary Roberts
Gary Roberts
Moderator
Joined: 9 Feb 05
Posts: 5842
Credit: 109408157894
RAC: 35253826

RE: ... the apparently

Quote:
... the apparently quieter "outer galaxy" beams compress much better than the "inner galaxy" ones we got recently.


OK, thanks for the explanation. I have a group of GPU endowed hosts at home where the plan I'm on has a modest 'peak time' allowance and a somewhat larger 'off-peak' allowance. Off-peak is 2.00am to noon and I arrange for all downloads to occur within that window. This makes it easy to look at the download action. The nice thing about the off-peak allowance is that if you exceed it, there are no extra charges but the download rate is throttled for the remainder of the month. This doesn't bother me as there's always enough time to finish the downloads by noon, even when downloading uncompressed data and throttled. I just need to make sure the downloading starts by about 5.00am and I was hoping to not have to monitor this quite so closely if the 'outer galaxy' rate of compression had been maintained.

Quote:
The network output of our download servers dropped to about a third (33%), which looks like it could well be the average of the compression ratios I've seen so far.


If the 'inner galaxy' data continues for a while, that will revert to much closer to 80-90%, I reckon. My daily downloads at home were around 7 to 8GB. After the initial change to compressed, I had a couple of days around 2 to 3GB and today it has jumped to about 5GB even though I actually lowered the cache size a bit once I saw the pattern of 1.8MB files. Now that I know the reason, I'll restore the cache size and expect to see 6GB or so as normal for me until the data changes again :-).

Cheers,
Gary.

mo.v
mo.v
Joined: 9 Jun 10
Posts: 13
Credit: 132471785
RAC: 0

I'm very pleased about this

I'm very pleased about this development. Yesterday 2 Feb I saw that BRP4 downloads were about 80 or 90% of their previous size, and new downloads today have been much smaller than that, probably averaging about 25% of last week's size.

I have a NVIDIA GeForce GTX 560 Ti (1023MB) driver: 30697 running on Intel i7 with Win7. During the last few days since the compression started I've had no download failures, no errors of any type and no invalid results. So I've noticed no problems so far.

Well done to the team.

Gary Roberts
Gary Roberts
Moderator
Joined: 9 Feb 05
Posts: 5842
Credit: 109408157894
RAC: 35253826

RE: ... Yesterday 2 Feb I

Quote:
... Yesterday 2 Feb I saw that BRP4 downloads were about 80 or 90% of their previous size, and new downloads today have been much smaller than that, probably averaging about 25% of last week's size.


Yes, we seem to be back on the more highly compressible "outer galaxy" data again, for the moment at least :-).

Cheers,
Gary.

Altair
Altair
Joined: 18 May 13
Posts: 6
Credit: 23171937
RAC: 0

Hey there, my Intel i5

Hey there,

my Intel i5 2500/ATI 6950 are doing this "BRPS (Perseus Arm Survey) 1.34 (opencl-ati)"-WU and till now it took already more then 36h, but hadnot even completed 45%.

Is it a problem with my setting(s) or take some of those WUs just that much time?

Thanks! ;)

Jord
Joined: 26 Jan 05
Posts: 2952
Credit: 5779100
RAC: 0

It's a problem with your

It's a problem with your settings. The Einstein OpenCL GPU applications like it best when they have a free CPU all for themselves to play with. Not when they have to fight for room on an overloaded CPU core.

So set your BOINC to use 3 CPU cores only, with the On multiprocessors, use 75% of the processors option in your computing preferences. (*) That way the one free CPU core will be able to help the GPU and you'll see your calculations speed up enormously. For comparison, my HD6850 runs BRP5s in approx. 17,500 seconds. Your HD6950 should be able to shave one- or two thousand seconds off of that when set up as above.

(*)When you use local advanced preferences, set it up there. These override the web preferences.

Altair
Altair
Joined: 18 May 13
Posts: 6
Credit: 23171937
RAC: 0

Hi Ageless, thanks a lot for

Hi Ageless,
thanks a lot for your help. It was just the way you said. Right after I decreased the available CPU-performance the calculations jumped up. It takes around 14,000s now. (It´s a 6970 BIOS-flashed 6950.)

Another question: How much of the capacity I took is used to support my GPU? And can my GPU after all exceed the lower CPU-performance or is the work just shifted from CPU to GPU?

Altair
Altair
Joined: 18 May 13
Posts: 6
Credit: 23171937
RAC: 0

RE: And can my GPU after

Quote:
And can my GPU after all exceed the lower CPU-performance or is the work just shifted from CPU to GPU?

Ok, was a bad question. After reading some other threads I can answer this by myself. ;)

Darren Peets
Darren Peets
Joined: 19 Nov 09
Posts: 37
Credit: 98000600
RAC: 30559

Bad workunit?

Bad workunit? http://einsteinathome.org/workunit/170250645

First invalid I've noticed in months, and all four tasks completed so far are marked as invalid.

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.