Limiting BRP4 to CUDA machines

Stranger7777
Stranger7777
Joined: 17 Mar 05
Posts: 436
Credit: 375,946,797
RAC: 145,451

RE: Have you donated enough

Quote:

Have you donated enough to pay for the additional relief machine? They do not have an unlimited budget, even if there's no heat problem.

Or would you prefer to donate only enough to buy a new cable for connecting to the relief machine they already have?

As I saw there, it is a possibility to search around and find enough components to make an additional machine, just by building one machine from 2 or 3 others broken ;)
I sometimes do the same in my office. Suddenly I understand that I have another machine to work on or to sell :)

Bernd Machenschalk
Bernd Machenschalk
Moderator
Administrator
Joined: 15 Oct 04
Posts: 3,985
Credit: 205,373,291
RAC: 36,199

The problem is that not any

The problem is that not any machine will do. The current 'relief' machine has 72GB RAM and six SSDs to handle the I/O load. We don't have that many such spare parts lying around.

BM

BM

Stranger7777
Stranger7777
Joined: 17 Mar 05
Posts: 436
Credit: 375,946,797
RAC: 145,451

RE: The problem is that not

Quote:

The problem is that not any machine will do. The current 'relief' machine has 72GB RAM and six SSDs to handle the I/O load. We don't have that many such spare parts lying around.

BM


This describes everything. This machine is a real monster for usual user :)
But it handles 6 WUGs simultaneously. What kind of SSDs do you use? Do they work in RAID? What size are them?

Bernd Machenschalk
Bernd Machenschalk
Moderator
Administrator
Joined: 15 Oct 04
Posts: 3,985
Credit: 205,373,291
RAC: 36,199

RE: This describes

Quote:
This describes everything. This machine is a real monster for usual user :)
But it handles 6 WUGs simultaneously. What kind of SSDs do you use? Do they work in RAID? What size are them?

6x 32GB SSDs, they were all that we could find spare. One for each WUG instance, a raid controller would reduce the IOps. The original WUG machine has 4x 250GB SSDs in a RAID, giving ~700GB. With the old version of the WUG this RAID (controller) became the bottleneck.

BM

PS: The new WUG will be tested today. If all goes well, it will go into production mode on Wednesday.

BM

Mike Hewson
Mike Hewson
Moderator
Joined: 1 Dec 05
Posts: 6,149
Credit: 129,261,843
RAC: 12,658

BTW Bernd, I haven't said how

BTW Bernd, I haven't said how much I like your new avatar! It neatly portrays your role in the project : "I can rig a force field box that'll hold the villi suspended in the centre. Then I can take it into the antimatter nacelle, put it into the regenerating chamber and release the forcefield by remote control" ( One Of Our Planets Is Missing, 1966 ) ...... :-) :-)

Cheers, Mike.

I have made this letter longer than usual because I lack the time to make it shorter. Blaise Pascal

joe areeda
joe areeda
Joined: 13 Dec 10
Posts: 285
Credit: 320,152,643
RAC: 23,016

RE: BTW Bernd, I haven't

Quote:

BTW Bernd, I haven't said how much I like your new avatar! It neatly portrays your role in the project : "I can rig a force field box that'll hold the villi suspended in the centre. Then I can take it into the antimatter nacelle, put it into the regenerating chamber and release the forcefield by remote control" ( One Of Our Planets Is Missing, 1966 ) ...... :-) :-)

Cheers, Mike.


I too like the avatar but I was reminded of a line from a ST parody which I don't remember the name of.

"Damn it Jim! I be an engineer, not a proctologist"

Stranger7777
Stranger7777
Joined: 17 Mar 05
Posts: 436
Credit: 375,946,797
RAC: 145,451

RE: BTW Bernd, I haven't

Quote:

BTW Bernd, I haven't said how much I like your new avatar! It neatly portrays your role in the project : "I can rig a force field box that'll hold the villi suspended in the centre. Then I can take it into the antimatter nacelle, put it into the regenerating chamber and release the forcefield by remote control" ( One Of Our Planets Is Missing, 1966 ) ...... :-) :-)

Cheers, Mike.

A little bit off topic, guys. But I'll close my eyes for the first time :)

2 Bernd. So you want to tell that RAID controller (is it on Intel chip and on external board?) has lower IO capabilities than direct connection, do you?

I'm using SSDs for about a year and found some disadvantages of them, especially in RAIDs. So I decided to use a software RAID instead.

Christoph
Christoph
Joined: 25 Aug 05
Posts: 41
Credit: 5,954,206
RAC: 0

It sounds like the IO load is

It sounds like the IO load is very high, so the question is, how much CPU would the process of the software raid use? Maybe it would still slow down production.

Greetings, Christoph

Stranger7777
Stranger7777
Joined: 17 Mar 05
Posts: 436
Credit: 375,946,797
RAC: 145,451

RE: It sounds like the IO

Quote:
It sounds like the IO load is very high, so the question is, how much CPU would the process of the software raid use? Maybe it would still slow down production.


It highly depends on the manufacturer. I'm using some Windows software RAID machines and some Intel ICH hardware RAID machines as database servers. I found that there is no difference in CPU consumption even while rebuilding (software RAID works on UDMA interface). Hardware RAID works a little bit faster, but not so stable and suitable as software one, especially when creating or killing RAID (hardware one each time kills all the data on both or more drives while software can use existing partition for that without any loss of data), changing disks etc. Besides that some recent hardware failures led hardware RAID to loss of data while no physical damage was made to drives. And it becomes even more actual when using SSDs with significantly faster speeds. 2 of these failures (each on hardware RAID) led to file system corruption and even to file zapping (disks were OK, it is only RAID malfunction). On the same time software RAIDs (all that I manage) survive a lot of disasters and have no any troubles yet.

Ok. Let's go back to question. Yes, software RAID still uses a little bit more CPU resources, but can not lead to CPU being stuck in IO ops. The bottleneck of the WUG machine I suppose is speed of SSD drives. And it cannot be made faster this time.

Stranger7777
Stranger7777
Joined: 17 Mar 05
Posts: 436
Credit: 375,946,797
RAC: 145,451

Looks like new BRP4

Looks like new BRP4 generation scheme works. At least now, cache tends to grow up. Good job guys.

Comment viewing options

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