Thread 'Best way to manage a fleet of machines?'

Message boards : Questions and problems : Best way to manage a fleet of machines?
Message board moderation

To post messages, you must log in.

AuthorMessage
greatext

Send message
Joined: 4 Dec 14
Posts: 4
Message 58422 - Posted: 4 Dec 2014, 10:06:45 UTC
Last modified: 4 Dec 2014, 10:13:19 UTC

Hey Boinc community,

I have been a long-time user of BOINC and want to promote its use through my job as an IT consultant. A number of clients of mine have expressed interest in loading it on all their machines, so I could put about 100 machines online tomorrow with that number doubling every 3-6 months or so.

I've been reading up on the wiki about weak account keys, account managers, etc and I'm a bit overwhelmed by all the choices. Can somebody suggest what might be the best way to accomplish this?

Here's what I need:
- Be able to create simple installer/config file/something that makes deploying this on many computers fairly painless.
- Be able to assign/remove projects and CPU/memory usage requirements update the boinc client automatically, and other various settings to all computers at once. I won't be manually updating them all. I need to set "per-project" CPU time/contribution limits. BAM for this? Bonus points if I can make groups of machines with different settings.
- Set up the clients in such a way that the person on the computer can't break into my account using it. Weak account keys look good, but I wouldn't be able to change my password and continue managing the machine?
- Not absolutely required, but I would like to be able to track stats of my "boinc fleet" and per individual machine/group of machines (so I can show my clients their computational contributions, they would love that)
- None of this can require port forwarding, remote access, a Windows server domain, etc

Also, on an unrelated note, since it's been a while since I looked into this: What's the stability of GPU support in BOINC? Obviously, not all cards are supported, but would rolling out GPU support to a fleet of machines potentially lead to instability, crashes, etc or are the kinks pretty much ironed out? If not, is there a quick way to test this on a per-machine basis?

And on another unrelated note, as a way to pitch this to my clients who are on the fence, has somebody compiled a master list of scientific discoveries/cool things discovered with BOINC? I remember coming across a list like this some time ago but couldn't seem to find it despite extensive googling.

Thanks for your help : )
ID: 58422 · Report as offensive
Coleslaw
Avatar

Send message
Joined: 23 Feb 12
Posts: 198
United States
Message 58423 - Posted: 4 Dec 2014, 14:04:57 UTC - in response to Message 58422.  
Last modified: 4 Dec 2014, 14:08:14 UTC

I can't answer all of your questions, but here is my take on a few of them.

I would not suggest setting up BOINC to use GPU's of systems that don't belong to you and aren't managed by you. The reason being is that every project codes their own apps which may affect the user vastly differently. If they have top of the line video cards, they may not experience the issues as much as a low end entry level card. Sometimes the generation of card will make a huge difference. The most important thing though is that most out of the box setups are not designed for the amount of heat running a GPU creates long term. So, my suggestion would be to stick with CPU only on such systems.

The simplest way to guarantee the GPU doesn't get used is to install the BOINC client as a service. That will also make sure it runs even if nobody is logged into the system. There is also an option during installation on whether to allow all users to control BOINC. Just uncheck that.

As far as the account key goes, I would prefer that you encourage them to have their own account so that the points/credit goes to them. That way you don't have to worry about your account at all. They may not want to bounce between projects but this may be the enticement needed to get more people interested. Simply putting on their machines means they will occasionally notice the blips that happen with projects and may be discouraged and tell you to pull BOINC off. If they are invested in it more, they are more likely to tolerate or ignore the random blip. I recommend a stable well respected project like WCG for something like this. It is easier to sell most people on finding cures than finding aliens.

As far as managing things, BAM is probably what you want. You can create groups and see all sorts of stats per machine and such. Get familiar with BOINCStats layout and I think that will be your best option.

I don't know of a master list of discoveries. Each project has their own policies on announcing findings and documentation. I think you will find WCG to be laid out well for some of these and has actually posted real world results lately rather than just another paper for someone else to follow up on.
ID: 58423 · Report as offensive
greatext

Send message
Joined: 4 Dec 14
Posts: 4
Message 58444 - Posted: 6 Dec 2014, 0:01:31 UTC

Thanks for the summary on GPUs, that's how I figured the current state was and definitely heat management could be an issue. For the CPUs, I'm going to hard-limit boinc to say 20-50% of the CPU so as to not increase power consumption much and leave wiggle room if their computer has heating problems (don't want this compounding issues a computer already has). I'll have to do some testing to find the magic numbers there.


I think my clients having their own accounts would be fine, which covers "part 1" of my deployment plan. In reality, I will have two types of deployments. The first one is to my business clients who have 5-20 machines which are left on 24/7 just begging to be BOINCed. For those, that is a good system. The other part of the plan is to install it on the computers of my residential repair clients. When they drop off their computers, they can check a box on the form that says "Install this awesome program" and then we'll handle setting it up, attaching it to projects etc. In this case, a different account for each client would be too burdensome to manage and set up.

For weak account keys, if I change the password, I would need access to the computer to update it on their end right? Is there an easier way to do that?
ID: 58444 · Report as offensive

Message boards : Questions and problems : Best way to manage a fleet of machines?

Copyright © 2025 University of California.
Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.