Information and Checklists for New Volunteers or for People rejoining after a period of inactivity

Gary Roberts
Gary Roberts
Moderator
Joined: 9 Feb 05
Posts: 4253
Credit: 12750314188
RAC: 25318964
Topic 198467

If you are not really familiar with the expanding range of exciting searches that run here at Einstein, here are some suggestions on what to do to make your initial startup a smoother and potentially less traumatic experience. If you already understand all the options and can plan for yourself, that's completely fine as these notes are not intended for you.

The suggestions below are just for having a better initial experience. Once you know how things work, feel free to set your preferences exactly as you want them.

I intend to edit this initial message to make corrections and add suggestions contributed by others. I will try to keep the list relevant for new/inexperienced volunteers. I hope to have the options explained clearly and kept as simple/non-technical as possible. This is NOT the thread for anything overly technical. Whenever this message is edited, I will flag the sentence, paragraph or section with a coloured symbol and date, using the following scheme #27-Feb-16

#27-Feb-16 -- for brand new information added.
#27-Feb-16 -- for information significantly modified for the sake of clarity, completeness or changed circumstances.
#27-Feb-16 -- for corrections where there was a minor error in the previous statement.
#27-Feb-16 -- for corrections where there was a significant error in the previous statement.

The Available Science Searches (Runs) in a Nutshell -- as of 25 Feb 2016

  • * The search for gravity waves (GW) - O1AS20-100T. This search covers the whole sky and is looking for continuous GW of the type that might come from spinning massive objects (eg neutron stars). These continuous emissions will be much more difficult to detect than

the 0.25 sec strong emissions from a merger of 2 black holes that was announced by the LIGO consortium recently. Our search is computationally intensive - perfect for a big 'supercomputer' like Einstein@Home. At present, this search will run only on CPU cores.

* The search for pulsars in binary systems that emit in the gamma-ray spectrum - FGRPB1. This search uses data from the large area telescope (LAT) that is on NASA's Fermi satellite. It is a follow-on from the FGRP4 search which was very successful in finding 13 brand new gamma-ray pulsars. The new emphasis is on finding pulsars in binary systems that emit gamma rays. At present, this search will run only on CPU cores.

* The search for radio pulsars in the southern sky - BRP6. This search uses data from the Parkes radio telescope in Australia. It covers parts of the cosmos that can't be 'seen' from the larger Arecibo telescope in Puerto Rico. There have been a large number of new discoveries by the analysis/re-analysis of this data because of the increasing sophistication of the search algorithms used here. This search is run on GPUs with a relatively minor amount of support from the CPU. There is plenty of data to process.

* The search for radio pulsars visible from Arecibo - BRP4, BRP4G. These searches consume the data faster than it can be generated so the discrete GPU version (BRP4G) will have periods of no work while waiting for new data. There is enough data for the much slower low power/mobile devices (such as Raspberry Pi and devices with ARM CPUs running Linux or Android) and Intel GPUs, which can be supplied through the BRP4 version of the run. If you wish to use any of these low power devices, check the applications page to make sure there is a suitable app for your hardware.#27-Feb-16

If you wish to volunteer for any of these searches, here is a summary of things to attend to before you add this project to the list. These notes assume that you have installed BOINC on your computer and that you may already have added other projects. You will need to look elsewhere if you need instructions on how to install BOINC. The comments below are designed to cover things to check so as to limit the initial range of science runs and numbers of tasks your host acquires when it first joins the project. Once you have things working steadily with a few returned tasks to provide some 'guidance' for BOINC to estimate run times, you can consider 'widening the scope' if you so desire.

Preference settings.

You should make sure you are familiar with the full range of settings available through links on your account page on the website. The most important ones are under "Computing preferences" and "Einstein@Home preferences". You should keep in mind the impact of 'website' preferences and 'local' preferences. If you set some values locally through BOINC Manager, they will override the equivalent website setting. If you 'clear' local settings, the website values of those will then apply once more. As you check through all available settings, please ask specific questions if you don't understand what a setting is for.

Make sure you understand how work cache settings are designed. There are two particular values called Store at least X days of work and Store up to an additional Y days of work. The first of these is a 'low water mark'. BOINC should always keep this amount of work. The 1st plus 2nd is a 'high water mark'. When BOINC does a work fetch it will fill up to this value. After that it shouldn't attempt to get more work until what's left is below the low water mark. Personally, I can't see why you would want two widely separated values UNLESS your internet connection is so limited that you need to take a 'big drink' quite infrequently. I much prefer to keep a constant amount of work on hand that would be sufficient to last through a short to medium term outage. For most people I would recommend making the 1st value as your actual setting and have a very low or zero value for the 2nd.

When you first join, the default is that all science runs are enabled in your project preferences. I think it will make things easier for you if you turn off everything and just enable a small subset. It makes things easier to manage if you choose a single 'CPU type' science run (either O1AST or FGRPB1) to start with. If you also have a suitable working GPU (and you want to use it), you can add one of the 'GPU' radio pulsar runs. I would recommend BRP6 to start with, simply because of the continuous availability of work. You should remove the check mark for the other projects until such time as you have enough experience to understand how to manage a bigger mix of science runs.

Another point to remember. BOINC requests new work based on your settings. Projects do not force extra work beyond what your settings allow. Also, it is BOINC that decides which tasks should be run and in what order. Any one project has no control over when BOINC decides to run its work.

Until BOINC knows better, it will be using project supplied 'estimates' for how long a task might take. In the initial stages, these estimates can be quite wrong. Make sure you plan for this possibility by restricting the work fetch until you have several completed results and the estimates are a lot closer to reality. BOINC will better manage a mix of projects if the work cache setting is kept well below the shortest task deadline. The more projects you have, the harder this all becomes.

Local vs Website Preferences

  • * On the BOINC Manager I use (7.2.42), 'computing preferences' are on the Tools menu item (Advanced view).
    * Not all settings can be changed locally.
    * Website preferences are good if you wish for a number of computers to have the same settings.
    * Changes in website preferences may take a while to propagate - depends on when your client next contacts the servers.

#27-Feb-16
* You can force a contact at any time by 'update'ing the project in BOINC Manager.#27-Feb-16
* If you have multiple projects, you should preferably choose and make all website changes on the one (master) website.
* Your most active project may be the best choice for that role.
* If you make global changes (not project specific) at different websites, the last set of changes will be those that apply.
* Local preferences are good if you want the change to apply only to the machine on which the change is made.
* Local preference changes take effect immediately and don't need an 'update'.

Some Specific Notes about Discrete GPU Crunching#27-Feb-16

  • * Current GPU apps are quite mature and well optimised. However, they do need a little CPU 'support'.
    * You don't have to have a 'fast' CPU to give that support. The support just needs to be 'available' when needed.
    * NVIDIA GPUs need less support than AMD GPUs and quite often no changes need to be made when running NVIDIA GPU tasks.
    * AMD GPUs need more support and 'medium to high end' GPUs can run incredibly slowly if that support is not immediately available.
    * The best way to provide support is to ensure that not all CPU cores are 'tied up' running CPU (or other intensive) tasks.
    * One convenient way to provide support is to change computing preferences to NOT allow BOINC to use 100% of cores. To use 3 out of 4, choose 75% for that setting.
    * It is often possible to improve output by running concurrent GPU tasks on the one GPU.

You should not attempt to do this until you really know what you are doing.
* If you run 2 concurrent tasks on an AMD GPU, you will automatically have one less CPU core being used by BOINC. This can often be quite satisfactory for GPU performance.
* It's impossible to really know what's best for your setup without you personally doing the experiments. The results can be quite rewarding.
If you decide to run multiple concurrent GPU tasks on certain high end AMD GPUs, (read here and read here, as just two examples), you will need to research which particular types give invalid results at any setting higher than single GPU tasks. The issue is probably driver related and I'm not aware that any solution has yet been found. There is no invalid results problem with any of these GPUs when running tasks singly.#27-Feb-16

Final Checks before Liftoff

This assumes BOINC is running on your computer and you are about to add the Einstein project.

  • * Check your work cache setting is 0.5 - 1 day maximum - less is better for starters.
    * Check you have selected the science run(s) you wish to start with and have removed the check mark on the others.
    * If you are doing GPU tasks, check you have 'allowed' the particular type(s) of GPU that you want to use.
    * If you have other projects, check the resource share that is set for E@H is appropriate (don't set extreme values).
    * If you don't understand a particular setting, leave it at the default.
    * Running BOINC Manager in 'Advanced view' makes it easier to see everything that is going on. Become familiar with what's on each tab and under each menu item.
    * The event log (on the Advanced menu) gives a summary of what BOINC is doing. The startup messages can be very useful for problem solving.
    * If you change any setting on the website, you can use the 'update' button on the Projects tab in BOINC Manager if you need BOINC to become immediately aware of the change.

I hope the above is of some use to new arrivals. I have probably forgotten lots of things so I welcome corrections/suggestions.

Cheers,
Gary.

Mike Hewson
Mike Hewson
Moderator
Joined: 1 Dec 05
Posts: 5798
Credit: 46450772
RAC: 2111

Information and Checklists for New Volunteers or for People rejo

Quote:
I hope the above is of some use to new arrivals.


.... and an older fossil too ! :-)

If I may ask one question ( but probably trigger others ). I've always got a bit muddled about the CUDA business for graphics cards. My belief is that CUDA is essentially proprietary to NVDIA. So currently for E@H :

- are there any limitations for user's choice of video technology ( not the card provider but the GPU manufacturer )? Or if you like what are, if any, the restrictions of E@H work units to be provided to a host given that a GPU is present at all ?

{ I hope that makes sense .... }

Cheers, Mike.

pascal_sig.jpg

Gary Roberts
Gary Roberts
Moderator
Joined: 9 Feb 05
Posts: 4253
Credit: 12750314188
RAC: 25318964

RE: I've always got a bit

Quote:
I've always got a bit muddled about the CUDA business for graphics cards.


Me too :-).

Quote:
My belief is that CUDA is essentially proprietary to NVDIA.


Yes. I have no idea if NVIDIA licenses that technology to other chip makers.

Quote:
... are there any limitations for user's choice of video technology ( not the card provider but the GPU manufacturer )?


To make sure this is clear, card provider == companies like Gigabyte, Asus, MSI, etc who use NVIDIA chips, whilst GPU manufacturer == NVIDIA or perhaps other companies that may enter into some sort of a deal with NVIDIA to fabricate the chips. None of this should be of any concern to a buyer of a CUDA capable GPU card other than perhaps a basic check to ensure the card is CUDA capable. Unless the card was extremely old or extremely low end, I imagine it would be. I must confess that apart from needing a minimum of 512MB of RAM (I think that still applies) I don't know if there is some lower limit to the version of the CUDA capability built into the GPU chip. CUDA was first introduced in 2006 and I'm sure I've seen GPUs of that vintage still crunching (extremely slowly) current Einstein tasks.

The competing technology is OpenCL as used for AMD and Intel GPUs and this thread is not the place for discussion about which one of these is 'better' or exactly what product a new user should buy into. The key message is that the appropriate GPU apps are available for automatic selection and download for a wide variety of different GPUs, either CUDA or OpenCL capable. Once the BOINC client on a user's computer detects the GPU and its capabilities and communicates that information to the Einstein servers, the appropriate app will be downloaded.

Quote:
Or if you like what are, if any, the restrictions of E@H work units to be provided to a host given that a GPU is present at all ?


Apart from 512MB, the only restriction I know of is that the earliest versions of OpenCL (1.0 I think) are not capable of supporting the OpenCL versions of Einstein GPU apps. I don't know of any similar restriction for CUDA. I'm not really sure I'm answering what you're asking :-). I wouldn't regard myself as having any real technical knowledge of these technologies.

Cheers,
Gary.

MarkJ
MarkJ
Joined: 28 Feb 08
Posts: 354
Credit: 48498004
RAC: 27351

You might want to add the

You might want to add the Raspberry Pi/Pi2 and Parallella can run BRP4 work on their ARM CPU. BRP4 work is not just limited to the Intel GPU. Although you did mention "low power/mobile devices" which they could be classified as.

Tiers Jean-Francois
Tiers Jean-Francois
Joined: 24 Nov 11
Posts: 60
Credit: 2057002
RAC: 0

Thanks for all these quite

Thanks for all these quite clear infos, Gary. Just the question below :

Quote:
The search for pulsars in binary systems that emit in the gamma-ray spectrum - FGRPB1 [...] At present, this search will run only on CPU cores.


Do you intend to make GPU's usable in the near future ? (On my small computer, even allowing the max CPU, LAT tasks take around 34 hours of CPU time, that is often more than 2 days in real time)
Cheers
JF

Gary Roberts
Gary Roberts
Moderator
Joined: 9 Feb 05
Posts: 4253
Credit: 12750314188
RAC: 25318964

RE: Do you intend to make

Quote:
Do you intend to make GPU's usable in the near future ? (On my small computer, even allowing the max CPU, LAT tasks take around 34 hours of CPU time, that is often more than 2 days in real time)


The most 'visible' spokesperson for things like this is Bernd. I'm just a volunteer like you so all I know is from what I read. There is a long running thread in the Science forum about future plans for Einstein. This is where Bernd tends to give insights about these sorts of things. He did make some comments about the possibility of new GPU apps not too long ago so I just went and had a look for that message. Here it is.

As for LAT tasks, have you made some changes along the lines I suggested in your thread over on the Problems ... board? Your most recent two LAT tasks had elapsed times of 57,436 and 156,324. I expected some sort of improvement but this is insane :-). This is not the thread for this discussion. Please take any questions or comments to your original thread.

Cheers,
Gary.

Gary Roberts
Gary Roberts
Moderator
Joined: 9 Feb 05
Posts: 4253
Credit: 12750314188
RAC: 25318964

RE: ... you did mention

Quote:
... you did mention "low power/mobile devices" which they could be classified as.


That was my intention, but I've expanded the content for BRP4 to more specifically mention your suggestions. I really know nothing about these devices. I assume if someone can get BOINC running on these they should be competent enough to verify that a suitable app exists before they waste too much time trying to get work.

Cheers,
Gary.

Gary Roberts
Gary Roberts
Moderator
Joined: 9 Feb 05
Posts: 4253
Credit: 12750314188
RAC: 25318964

I've made some additions to

I've made some additions to the original message. In particular, I've added a complete new section containing notes about discrete GPU crunching. I've also added a system of colour indications to highlight those parts that have been added or amended for the benefit of anybody trying to decide what's new and what's not.

I've deliberately avoided Intel GPUs because (running Linux) there weren't any Linux drivers last time I looked. This was a while ago so maybe there is something now in development. I don't have time to get involved until there is something that works, pretty much 'out of the box'.

Since I have no experience, I'd rather say nothing than something that might lead people astray. I haven't even been following those struggling to find a 'good' driver version for Windows. I just know it's a bit of a lottery :-).

Cheers,
Gary.

Snow Crash
Snow Crash
Joined: 24 Dec 09
Posts: 65
Credit: 100880785
RAC: 0

Thanks for putting this

Thanks for putting this together ... any thoughts on matching the actual preference settings with the "Available Science Searches"?
Maybe add clarification as to which are CPU / GPU / both?

Binary Radio Pulsar Search (Arecibo)
Binary Radio Pulsar Search (Arecibo, GPU)
Binary Radio Pulsar Search (Parkes PMPS XT)
Gravitational Wave search O1 all-sky F
Gravitational Wave search O1 all-sky I
Gamma-ray pulsar search #4
Gamma-ray pulsar binary search #1

--------------------------
- Crunch, Crunch, Crunch -
--------------------------

Cadeus
Cadeus
Joined: 1 Aug 05
Posts: 4
Credit: 15422306
RAC: 17532

What about minimum system

What about minimum system requirements?  A post in this section seems like a good place.

Einstein requirements are listed on the following page, but the last update was December of 2010.

http://boinc.berkeley.edu/wiki/List_of_projects_by_system_requirements

I like to re-purpose other peoples old computers to run BOINC projects. As the computer and OS get outdated, I have to drop projects. I suspect Einstein doesn't have a project for a P3 running linux. (It doubles as a heater in winter Smile )

 

Robert Klein
Robert Klein
Joined: 2 Dec 16
Posts: 7
Credit: 428808600
RAC: 1051233

But don't forget the very bad

But don't forget the very bad power-efficiency of real old hardware! I don't think, that for example a "Northwood"-CPU with a GeforceFX 5k would reach a justifiable level of power-efficiency.

Comment viewing options

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