Message boards : News : Universe@Home and BOINC Pentathlon on admin point of view
Message board moderation

To post messages, you must log in.

AuthorMessage
Profile Krzysztof Piszczek - wspieram ...
Project administrator
Project developer
Project tester
Avatar

Send message
Joined: 4 Feb 15
Posts: 841
Credit: 144,180,465
RAC: 2
Message 1240 - Posted: 27 Jun 2016, 18:22:04 UTC
Last modified: 27 Jun 2016, 18:24:54 UTC

Better late than never, so a few words about how Pentathlon looks from the
administrator’s point of view

As I have explained before, loads of pressure hits the server with thousands of
computers connecting with "bunkered" WU's. Usually we had ~45'000-50'000 tasks
in progress and just before the start the quantity grew up to 250'000 and later
stayed at the level of about 220'000-240'000 tasks in progress.

As you probably noticed from the picture showed in the other topic the only
problem we had was with hard disk speed because of thousands database operations
in the background of other server tasks.

As you can see in the picture below, almost in the whole Pentathlon competition
time disk utilisation was at 100% or just below it. We usually have stable disk
utilisation at 40-50%.

In the middle of the competition I also switched off all unnecessary daemons and
other software which usually works on our server and even without them the
server was off-line for 30 minutes.

Now, the situation is stable. We still have more tasks in progress than before
the competition (which makes us very happy) and we will also prepare to buy some
SSD's asap :)

Thank you very much for your power in Pentathlon and obviously for every day
with us and our tasks :)


Lepiej późno niż wcale. Tak więc kilka słów o tym jak Pentathlon wyglądał z
punktu widzenia administratora

Jak tłumaczyłem wcześniej, bardzo duże obciążenie pojawia się na serwerze gdy
tysiące komputerów podłączą się ze swoimi "zbunkrowanymi" WU. Zwykle mamy
~45'000-50'000 zadań w toku, a zaraz przed rozpoczęciem ta ilość wzrosła do
250'000, a później pozostała na poziomie 220'000-240'000 zadań w toku.

Jak pewnie zauważyliście na załączonym obrazku, jedyny problem, jaki mieliśmy to
opóźnienia w operacjach dyskowych i wynikał on z tysięcy operacji na bazie danych.

Również widać na poniższym obrazku, że prawie przez cały czas trwania konkursu
Pentathlon wykorzystanie dysku było na poziomie 100% lub nieco poniżej. Zwykle
mamy wykorzystanie dysku na stabilnym poziomie 40-50%.

W środku konkursu wyłączyłem wszystkie zbędne demony i oprogramowanie zwykle
pracujące/działające na serwerze i nawet bez nich serwer był niedostępny przez
30 minut.

Teraz sytuacja jest stabilna. Nadal mamy więcej zadań w toku niż przed
konkursem (co bardzo nas cieszy) i będziemy się przygotowywać do kupienia dysków
SSD tak szybko jak to możliwe.

Dziękujemy bardzo za zaangażowanie w Pentathlon i oczywiście za każdy dzień z
nami i z naszymi zadaniami :)
Krzysztof 'krzyszp' Piszczek

Member of Radioactive@Home team
My Patreon profile
Universe@Home on YT
ID: 1240 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile yoyo_rkn

Send message
Joined: 4 Feb 15
Posts: 1
Credit: 3,464,111
RAC: 0
Message 1241 - Posted: 27 Jun 2016, 20:34:16 UTC - in response to Message 1240.  
Last modified: 27 Jun 2016, 20:35:55 UTC

You have to keep your Boinc DB very small. This drastical reduces disk load. Do it by following meassures:
    Create only less workunits. If ready to sent goes below a threshold (e.g. 1000), create new ones (e.g. another 1000).


    Delete workunits soon after they are validated/assimilated. During race time I delete them 4 hours after assimilation.


    Create workunits (if possible) which are running long (e.g. 7 hours)



Additional forbid frequent client requests. Allow clients to connect only every 10 minutes.
yoyo

ID: 1241 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Krzysztof Piszczek - wspieram ...
Project administrator
Project developer
Project tester
Avatar

Send message
Joined: 4 Feb 15
Posts: 841
Credit: 144,180,465
RAC: 2
Message 1242 - Posted: 28 Jun 2016, 7:17:27 UTC - in response to Message 1241.  

Thx Yoyo :)

Actually file_deleter delete files just after assimilation, db_purge keeps database in sensible size.
Work unit generator is separate topic. To keep work in right order I have to generate jobs in batches of 100'000 tasks (at the moment) and task duration isn't really my choice - it's based on application.
Krzysztof 'krzyszp' Piszczek

Member of Radioactive@Home team
My Patreon profile
Universe@Home on YT
ID: 1242 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Chooka
Avatar

Send message
Joined: 30 Jun 16
Posts: 41
Credit: 309,815,029
RAC: 0
Message 1249 - Posted: 8 Jul 2016, 7:55:11 UTC

If you need to buy anything for this project, please email Henri over at Bitcoin Utopia. As a participate of BU, we need more worthy projects to fund (like new SSD's) so please apply for a campaign!

ID: 1249 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote

Message boards : News : Universe@Home and BOINC Pentathlon on admin point of view




Copyright © 2024 Copernicus Astronomical Centre of the Polish Academy of Sciences
Project server and website managed by Krzysztof 'krzyszp' Piszczek