Российские распределенные вычисления на платформе BOINC
Форум участников распределённых вычислений.

Добро пожаловать, Гость! Чтобы использовать все возможности Вход или Регистрация.

Уведомление

Icon
Error

Опции
К последнему сообщению К первому непрочитанному
Offline cykodennis  
#1 Оставлено : 20 июля 2014 г. 17:09:07(UTC)
cykodennis


Статус: Я тут не впервой

Группы: Member
Зарегистрирован: 10.07.2014(UTC)
Сообщений: 21

Сказал(а) «Спасибо»: 1 раз
Поблагодарили: 6 раз в 4 постах
Conclusion first (for the quick readers):
Possible, but it sucks.


To be able to run Gerasim@home on a LINUX OS is a very understandable desire - at least for me.

You may have technical beliefs, political reasons, want to spare the money for the licenses or have just a legitimate portion of William-Gibson-Cyberpunk Paranoia - there are several very good reasons to avoid Microsoft Windows as your main OS.
(Some of them are even similar to the reasons you may have to explicit support russian DC projects - but we won't discuss this any further.)

Your problem is that G@H just provides you with an Windows application and you have to find a way how to come along with that.
Don't expect support by the project crew - you are doing all of this on own risk.

So, you have two possible ways to go - WINE or Virtualization.
While there are no doubts that running G@H on virtual machines will succeed very well, the question about the performance in WINE is still open.

So i did the test.

Wine 1.6.1
Scientific Linux 6.5
BOINC 7.2.28

First of all you have to install the Windows BOINC Client in Wine. You will succeed with exactly the flaws which are described here - no service installation f.e.
After that you may run in a problem that you have to connect to the client, because the automatic connect via Manager start has failed.
But than, you'll have success in connecting to G@H and downloading your first Workunits.

(i did not do any special configurations or tweaks, so following observations are strictly about a fresh BOINC install on Wine, out of the box)

Best things first - you will get valid results, so you are in general able to run G@H without Windows at all.
WU 3359115 f.e. was crunched via WINE by me.

But there are two big disadvantages:
- Performance

BOINC is running 4-5 times slower than a native installation on the same host. I noticed in the logs, that at start the benchmark estimates my CPU power ~25% of its real capacity. I don't have an explanation for that, but i can confirm that the running time of single WUs correlates nearly exactly with this wrong benchmark results.
Nonetheless my cores did run on full power (100%).

- BOINC "forgets" the CPU time, needed for completation of the task

You can watch that when stopping BOINC and starting again while crunching a G@H WU. While the counter for the percentage of the WU will still display correctly, the timecounters are reset to zero.
After completing the task, your BOINC tells the project server are crunching time of 0 (in words - "ZERO"wink - this could be an issue for you, because the CPU time is (often?) needed for calculating your credit for the WU.

I can't deny that there may be tricks or tweaks which could improve this performance - but i think it is not very probably.

So, my conclusion is:
Although it is possible to run G@H via WINE on LINUX, i don't recommended that.
The performance is awful and the issue with the CPU time may bring up more problems for you and maybe your wingmen.

P.S.:
In WINE HQ you can find following suggestion
"It should be possible to use a native BOINC and run a Windows executable with Wine if absolutely necessary."

I've tested that too, but it failed even at the beginning, because the native BOINC identifies itself to the project server as LINUX.
I guess I would have to edit some files of BOINC and of the project, but i didn't tested that yet.

You can get near native performance with the KVM-QEMU Virtualization Hypervisor and at least moderate performance with VirtualBox, so messing around with WINE and BOINC would have made only sense, when you could get superior performance without any editing in BOINC (in my point of view).

I will come up with a description of a simple KVM-QEMU setup for those LINUXers, who are not so familiar with the topic, in the next weeks.

Man's unfailing capacity to believe what he prefers to be true rather than what the evidence shows to be likely and possible has always astounded me.
Offline evatutin  
#2 Оставлено : 20 июля 2014 г. 22:46:36(UTC)
evatutin


Статус: Старожил

Медали: Первооткрывателю: Результат в проекте SAT@homeРазработчику: За организацию проекта Gerasim@home

Группы: Editors, Member
Зарегистрирован: 08.06.2010(UTC)
Сообщений: 3,359
Откуда: Russia, Kursk

Сказал(а) «Спасибо»: 931 раз
Поблагодарили: 1507 раз в 734 постах
Цитата:
BOINC "forgets" the CPU time, needed for completation of the task


Last version of the wrapper (spstarter 1.10+) must save CPU time after suspending/resuming computation, strange behavior...

kvt.kurskstu team founder
Gerasim@home scientist
My numbers are 5056994653507584 and 1835082219864832081920. Why not? smile
Offline cykodennis  
#3 Оставлено : 21 июля 2014 г. 0:18:08(UTC)
cykodennis


Статус: Я тут не впервой

Группы: Member
Зарегистрирован: 10.07.2014(UTC)
Сообщений: 21

Сказал(а) «Спасибо»: 1 раз
Поблагодарили: 6 раз в 4 постах
You think it is the behaviour of the wrapper?
My guess was, it would be anything wine-specific.

Has the knowledge about this any value for you? My test machine is still available, i could run another project to verify?
Man's unfailing capacity to believe what he prefers to be true rather than what the evidence shows to be likely and possible has always astounded me.
Offline evatutin  
#4 Оставлено : 21 июля 2014 г. 0:39:00(UTC)
evatutin


Статус: Старожил

Медали: Первооткрывателю: Результат в проекте SAT@homeРазработчику: За организацию проекта Gerasim@home

Группы: Editors, Member
Зарегистрирован: 08.06.2010(UTC)
Сообщений: 3,359
Откуда: Russia, Kursk

Сказал(а) «Спасибо»: 931 раз
Поблагодарили: 1507 раз в 734 постах
Автор: cykodennis Перейти к цитате
My test machine is still available, i could run another project to verify?


If it is interesting you can try some additional WUs from Gerasim. Not all another projects needs wrapper.

kvt.kurskstu team founder
Gerasim@home scientist
My numbers are 5056994653507584 and 1835082219864832081920. Why not? smile
Offline cykodennis  
#5 Оставлено : 21 июля 2014 г. 0:49:04(UTC)
cykodennis


Статус: Я тут не впервой

Группы: Member
Зарегистрирован: 10.07.2014(UTC)
Сообщений: 21

Сказал(а) «Спасибо»: 1 раз
Поблагодарили: 6 раз в 4 постах
I won't go any further on the WINE path, because it's overall no good solution for crunching purposes.
I guess, only the most desperate penguins will accept crunching times 4-5 times slower than normal.

My question was, if it is interesting for you. In this case i would try to check out if this strange behaviour is gerasim or wine specific.
Man's unfailing capacity to believe what he prefers to be true rather than what the evidence shows to be likely and possible has always astounded me.
Offline evatutin  
#6 Оставлено : 21 июля 2014 г. 9:55:57(UTC)
evatutin


Статус: Старожил

Медали: Первооткрывателю: Результат в проекте SAT@homeРазработчику: За организацию проекта Gerasim@home

Группы: Editors, Member
Зарегистрирован: 08.06.2010(UTC)
Сообщений: 3,359
Откуда: Russia, Kursk

Сказал(а) «Спасибо»: 931 раз
Поблагодарили: 1507 раз в 734 постах
I suspect that virtual machine is not so effective. Have you tried another projects at the same conditions?

kvt.kurskstu team founder
Gerasim@home scientist
My numbers are 5056994653507584 and 1835082219864832081920. Why not? smile
Offline cykodennis  
#7 Оставлено : 21 июля 2014 г. 10:50:46(UTC)
cykodennis


Статус: Я тут не впервой

Группы: Member
Зарегистрирован: 10.07.2014(UTC)
Сообщений: 21

Сказал(а) «Спасибо»: 1 раз
Поблагодарили: 6 раз в 4 постах
Автор: evatutin Перейти к цитате
I suspect that virtual machine is not so effective.


Well, that's fortunately not true anymore. You can get near native performance with the KVM-QEMU hypervisor, when your CPU supports hardware virtualization.
I've already a virtual machine running G@H, you can watch it here: Computer 19329

As you may notice, the measured floating point/integer speeds are not so bad for an Intel I5 (which is the CPU on the Hypervisorhost). Practically, the G@H WUs aren't crunched remarkably slower than on bare metal.
So, with that in mind, the WINE solution would only make sense, if it can reach the same performance with equal or less configuration efforts.


Автор: evatutin Перейти к цитате
Have you tried another projects at the same conditions?


No, i have not tried it yet. That was my question yesterday, if such a testing would have any value for the project.
But i will, using yoyo@home, which uses a wrapper too (AFAIK).

I'll tell you as soon as i have got some results.

Man's unfailing capacity to believe what he prefers to be true rather than what the evidence shows to be likely and possible has always astounded me.
Offline cykodennis  
#8 Оставлено : 22 июля 2014 г. 23:13:56(UTC)
cykodennis


Статус: Я тут не впервой

Группы: Member
Зарегистрирован: 10.07.2014(UTC)
Сообщений: 21

Сказал(а) «Спасибо»: 1 раз
Поблагодарили: 6 раз в 4 постах
I am actually crunching yoyo@home on the test machine, as promised, and i am seeing the first results.
Although the WUs behave in detail different, the overall conclusion remains the same. Yoyo WUs seem to not forget the CPU time completely after being stopped, but that affects only the counter of the BOINC GUI.

The CPU time claimed to the project server is ranging near zero again, so it seems to be a WINE specific issue.
I also found a few years old message board post, referring exact to this problem. So i think it's really something which has come with the development of WINE and BOINC.

The issue with the slowed down CPU (and the wrong benchmarks) has shown up again too. Maybe i've found an hint for that. While watching the system monitor of the host, I've noticed that all four cores are used at ~80% capacity, although only two cores are assigned.

However, i don't think that it really matters anymore. Running Windows Boinc via WINE was a performance boosting trick in good old times, today it is simply not needed anymore, with all those native clients and the progress in VM technology. So i don't expect any improvements.

I've started this thread for my fellow LINUX tovarishs, to check out if WINE is a solution for running G@H on Linux hosts. I think my first conclusion was the most helpful one:
Possible, but it sucks. smile

As promised, i will write a guide for setting up a nice VM enviroment for running G@H on LINUX. But it will take some time, i have become actually a little bit busy with other things.
Man's unfailing capacity to believe what he prefers to be true rather than what the evidence shows to be likely and possible has always astounded me.
thanks 1 пользователь поблагодарил cykodennis за этот пост.
evatutin оставлено 22.07.2014(UTC)
Пользователи, просматривающие эту тему
Guest (2)
Быстрый переход  
Вы не можете создавать новые темы в этом форуме.
Вы не можете отвечать в этом форуме.
Вы не можете удалять Ваши сообщения в этом форуме.
Вы не можете редактировать Ваши сообщения в этом форуме.
Вы не можете создавать опросы в этом форуме.
Вы не можете голосовать в этом форуме.

Boinc.ru theme. Boinc.ru
Форум YAF 2.1.1 | YAF © 2003-2017, Yet Another Forum.NET
Страница сгенерирована за 0.173 секунды.