Posts by Sergey Kovalchuk

1) Questions and Answers : Windows : Just one windows machine with errors (Message 722)
Posted 27 Oct 2020 by Sergey Kovalchuk
Post:
there is one more way to solve the problem of library compatibility - the preinstalled and configured PyTorch package
you need to choose a way to switch between the native and downloaded packages

the BURP project used a switch file - it was necessary to create a file or folder "native" in the project folder
but the "blender" still loaded with the project and took up space.

on another project, I already proposed another selector - on the configured host (with the PyTorch package) create a pseudo coprocessor of the same name and configure one version of the application (without libraries) to use it through the plan-class

<cc_config>
  <options>
    <coproc>
      <type>PyTorch</type>
      <count>1</count>
      <non_gpu/>
    </coproc>
  </options>
</cc_config>


<app_version>
    <app_name>mlds</app_name>
    <avg_ncpus>1.00000</avg_ncpus>
    <plan_class>PyTorch</plan_class>
    <coproc>
        <type>PyTorch</type>
        <count>1.000000</count>
    </coproc>
</app_version>
2) Questions and Answers : Issue Discussion : Test tasks not sent? (Message 677)
Posted 23 Oct 2020 by Sergey Kovalchuk
Post:
Is there any prioritization of tasks, such as re-released ones?
Shortened deadline? "Reliable" hosts for replication?
I.e. any options that can push rare test tasks out of the queue

It is worth trying to reissue test tasks - re-add them to the queue
3) Questions and Answers : Issue Discussion : Test tasks not sent? (Message 657)
Posted 17 Oct 2020 by Sergey Kovalchuk
Post:
maybe the problem will be solved by increasing the size of the shared memory
this error is common for many projects with default settings
appears after adding a second application with a large difference in the volume of tasks
it is also advisable to tweak the parameters of the scheduler queue - so that there is guaranteed space for each application
4) Message boards : News : Badges! (Message 640)
Posted 11 Oct 2020 by Sergey Kovalchuk
Post:
10K (onyx),

500K (amethyst),

1M (sapphire),

500M (ruby),

1B (emerald)
5) Questions and Answers : Unix/Linux : GPU update (Message 576)
Posted 4 Oct 2020 by Sergey Kovalchuk
Post:
You may be interested in the following information.
I didn’t go deep into ML methods, but the abbreviations seemed familiar to me. And I found them in the list of my "BOINC trophies"
The now virtually dead project Citizen Science Grid in 2017 launched different versions of image processing applications
(starting with manually searching and identifying birds in photos)
The names had similar abbreviations - CIFAR-10, MNIST, CNN, Scaled FMP, etc.
6) Message boards : News : Badges! (Message 562)
Posted 2 Oct 2020 by Sergey Kovalchuk
Post:
inconsistency with (test) - the application dropped out of statistics
on statistics page Participants - Per application - array size error
7) Message boards : News : Badges! (Message 561)
Posted 2 Oct 2020 by Sergey Kovalchuk
Post:
And hopefully these badges will be made available to the stats sites? (Milestone/special badges) :)

https://www.mlcathome.org/mlcathome/stats/tables.xml

badges are already in export data.
but it is better to exclude RAC badges from export
anyway, over time they will disappear from signatures

PS I would like to clarify
2 applications are formally defined - MLDG and test
Are they counted as one or the test only goes to total?

PS2 I missed the preliminary discussion, but I would like to give an opinion
accumulate better special badges with unique designs
linear ones of the same type (10k 100k 1M), especially with per_apps accounting,
it is better to replace the previous ones with a new one - that is, one badge per app
8) Questions and Answers : Issue Discussion : mldstest (Message 471)
Posted 12 Sep 2020 by Sergey Kovalchuk
Post:
I agree there should be the option to choose not to run this app in preferences, and I'm surprised its not there automatically. We're looking into it.


https://boinc.berkeley.edu/trac/wiki/AppFiltering

BOINC offers two mechanisms allowing users to control the set of apps for which they get jobs.
The first, lets you designate apps as "beta test", and lets users opt-in to getting jobs for these apps.
The second, "app filtering", lets users select apps individually.

App filtering is disabled by default. To enable it, add to your html/project/project_specific_prefs.inc: 
define ('APP_SELECT_PREFS', true);


The app filtering user interface is part of project-specific preferences (i.e., the "Preferences for this project" link on their account page).
If the feature is enabled, users are shown:
• A list of apps and check boxes.
• A checkbox labeled "If no work for selected applications is available, accept work from other applications?"
9) Questions and Answers : Issue Discussion : mldstest (Message 466)
Posted 11 Sep 2020 by Sergey Kovalchuk
Post:
a mldstest application appeared on the server
it is desirable to activate the choice in the settings
1 "run only selected applications" in preferences
2 mark the application or its version with the "beta test" flag
then the option "run beta test" will appear in preferences
10) Questions and Answers : Issue Discussion : News notifications to be displyed in BOINC manager, too (Message 304)
Posted 8 Aug 2020 by Sergey Kovalchuk
Post:
I almost wish there was a way to put a summary in the client news section that points to the whole post.

divided the news into two parts - a short announcement and the main text.
the announcement will form a topic in the news section, will be displayed on the home page of the site and in the client's notifications. there is always a "discussion" - a link to the topic in the forum
publish the main text as a response to the announcement - the second message in the topic
11) Questions and Answers : Issue Discussion : MLC@home WUs using 2 CPUs (Message 248)
Posted 25 Jul 2020 by Sergey Kovalchuk
Post:
I'm clearly missing something..

Max # jobs - 1..8, No limit - receive no more than # tasks per host, regardless of the queue size

just will allow you to avoid an irregular number of tasks to the detriment of the rest - but control in the user settings, in addition to hard server limits

- if the task duration is incorrectly determined (bad benchmark)
- common queue of several projects without overflow on the first request
- for example, only one task on host with 4 cpu and 2G memory, no need "to be quick" to stop the host before the crash
12) Questions and Answers : Unix/Linux : /lib64/libc.so.6: version `GLIBC_2.18' not found (Message 222)
Posted 22 Jul 2020 by Sergey Kovalchuk
Post:
Application does not work on CentOS 7.

https://www.mlcathome.org/mlcathome/forum_thread.php?id=5&postid=76
13) Questions and Answers : Issue Discussion : MLC@home WUs using 2 CPUs (Message 205)
Posted 22 Jul 2020 by Sergey Kovalchuk
Post:
I'm not sure if there's something that can be done about this from your end?

from server side can allow extended user preferences

Max # jobs - 1..8, No limit
Max # CPUs - 1..8, No limit
14) Message boards : Cafe : Badges (Message 185)
Posted 19 Jul 2020 by Sergey Kovalchuk
Post:
I could be convinced to bump it to 10K, but that's about as high as I'm willing to go.

well, create everything in the line 1k 10k 100k 1M 10M 100M 1G 10G 100G
1st - start, closed black box ;-)
the last one - for growth, top RAC users will get it quickly

I to am worried about how these look in 48x48 though.

on home page - 56 * 56, list of users 24 * 24 and forum 20 * 20
15) Message boards : Cafe : MLDS Live Status (Message 181)
Posted 18 Jul 2020 by Sergey Kovalchuk
Post:
the "MLDS Live Status" table can be built in the project itself
assign a batch number to tasks of each type
on the page server status after the table tasks by application add tasks by batch
example
16) Message boards : Cafe : Badges (Message 180)
Posted 18 Jul 2020 by Sergey Kovalchuk
Post:
I'm thinking these ranking
Top 5/10/25% RAC: gold, silver, bronze
1K credits, 1M credits, 1B credits: red/green/blue
And one special "Early Adapoter" badge (EA) for anyone who receives credit by July 30th.

IMHO, it is better to leave standard images for the top 1-5-25% and exclude them from export
they are useless for signatures and statistics sites

achievement badges are too complex, when scaled to 32 * 32 they will lose their expressiveness
you can leave 1M and MLC on a simple color field - a circle, hexagon or shield

the first badge 1k is too small (4 WU), but 100k is too much for the entry level (this is now Top100), I'm for 10k and more gradations

PS if you plan to add other applications / subprojects later, then you can enable "by application statistics" and add badges for achievements separately in each subproject, then instead of the MLC you need MLDS
17) Questions and Answers : Issue Discussion : misconfigured site (Message 168)
Posted 12 Jul 2020 by Sergey Kovalchuk
Post:
https://www.mlcathome.org/mlcathome/index.php
<!-- <scheduler>https://www.mlcathome.org/mlcathome_cgi/cgi</scheduler> -->
<link rel="boinc_scheduler" href="https://www.mldsathome.org/mlcathome_cgi/cgi">
<title>MLC@Home</title>
18) Questions and Answers : Issue Discussion : You need a "Number Crunching" forum (Message 162)
Posted 12 Jul 2020 by Sergey Kovalchuk
Post: 
Tasks in progress	28723
Computers with recent credit 518

it is possible to limit the number of tasks per host (per CPU core)
19) Questions and Answers : Windows : Windows app up for testing! Report issues here (Message 137)
Posted 8 Jul 2020 by Sergey Kovalchuk
Post:
check if the VC 2015 (x64) redistributable runtime is installed in the system.
probably libraries coming with the application are not compatible with your system
20) Questions and Answers : Windows : Thoughts on windows requirements? (Message 111)
Posted 6 Jul 2020 by Sergey Kovalchuk
Post:
You are not the first project compiled under VS2019. Active use of BOINC collects a whole collection of redist from 2005 to 2015.
But it is imperative to indicate and highlight in the announcement the requirement of a specific version - 2015 and if both architectures are needed (x32 and x64).
For 2015, it’s better to indicate the correct link, for example Microsoft Visual C++ Redistributable for Visual Studio 2015, 2017 and 2019


By the way, under Linux, you also sometimes have to install additional packages.
It would be nice to consider the possibility of using native Pytorch for the client distribution.
If this is not too complicated installation procedure.
At least one project has used this way to resolve the library dependency conflict


Next 20

©2022 MLC@Home Team
A project of the Cognition, Robotics, and Learning (CORAL) Lab at the University of Maryland, Baltimore County (UMBC)