Rebuilding the farm

Message boards : Number crunching : Rebuilding the farm
Message board moderation

To post messages, you must log in.

1 · 2 · Next

AuthorMessage
Jason Ebacher

Send message
Joined: 16 Aug 99
Posts: 19
Credit: 5,692
RAC: 0
United States
Message 5941 - Posted: 10 Jul 2004, 4:18:08 UTC

Ok I have a small crunching farm up and was doing quite well, them we moved and i recently got most of it back online... I was and still am running the classic seti app, had one "server" with folders setup for each node where everything was stored and ran "bartPE" as my OS, (WINDOWS XP on a CDROM, no harddrive needed, only requires a MOBO CPU RAM CDROM drive and NIC) for the nodes that looked to this share on the network for the client and the work to do.. I then used seti watch to monitor my progress on my normal pc... I also Created a Linux floppy that downloads the CLI and crunches them... (for the hardware tha was not up to par for the BARTPE OS.

SO now with the new app it requires a install and throws DLLs all over the place ect... and no way to centrally locate all the data and a central monitoring system.. (or am I wrong)

SO what are others that are dong farms, doing for reporting storage and farm systems in general.

If I have to redo my farm I want to start now and get a good setup before I convert all the PC's over.

Thanks,
Jason

ID: 5941 · Report as offensive
Jason Ebacher

Send message
Joined: 16 Aug 99
Posts: 19
Credit: 5,692
RAC: 0
United States
Message 6581 - Posted: 11 Jul 2004, 17:43:38 UTC
Last modified: 11 Jul 2004, 17:44:03 UTC

No One, Hmm..., no one has converted there farm over.. and needed reporting monitoring tools?

Jason
ID: 6581 · Report as offensive
Profile RampageIII
Avatar

Send message
Joined: 23 Oct 01
Posts: 9
Credit: 749,981
RAC: 0
United States
Message 6589 - Posted: 11 Jul 2004, 18:04:01 UTC
Last modified: 11 Jul 2004, 18:04:13 UTC

This program is an add-on for the BOINC project, similar to the SETI@home I tool called SetiWatch. It can watch all BOINC clients throughout a network and lists all informations about these clients. The program can also log BOINC specific results.


BoincView


ID: 6589 · Report as offensive
Darren
Volunteer tester
Avatar

Send message
Joined: 2 Jul 99
Posts: 259
Credit: 280,503
RAC: 0
United States
Message 6590 - Posted: 11 Jul 2004, 18:15:29 UTC - in response to Message 6581.  

> No One, Hmm..., no one has converted there farm over.. and needed reporting
> monitoring tools?

I think it's just that it's so early in the process right now that most people with large farms are still trying to figure out the best way to do things.

In my case, I only have 7 systems running now (had 9, retired 2 recently). With seti 1 (and folding@home), my farm units were booting from floppy using trinux, then working in their directories on my server. Boinc will not run under trinux, so I put harddrives back on all my farm units to get a full os for each one. I don't think your os will be as big a problem as mine was, so it shouldn't require anything as drastic as that.

As for monitoring, that's definately limited at the moment. All of my dealings are with linux, and one of the 3rd party software creator sites says at some point they will implement multi-system monitoring via samba in their linux monitor, but it's not there yet. In the meantime, I just created multiple desktops on my server (one for each remote system) and start the monitor software from the remote system on the server itself via ssh. This starts an instance of the 3rd party gui from the remote system that displays on the local system. Then I can just click through the desktops to each system and see the monitoring software for that system.

Definately a crude and unrefined way to do it, but with my small setup it works ok.



ID: 6590 · Report as offensive
ric
Volunteer tester
Avatar

Send message
Joined: 16 Jun 03
Posts: 482
Credit: 666,047
RAC: 0
Switzerland
Message 6620 - Posted: 11 Jul 2004, 19:47:48 UTC

5 members of my farm crunshed under Linux, the LPR (Linuy Router Project)
boot disk where not easy to setup, but time done, stable as a rock.
-1-
put back HD (for the moment) :-(
but I will try to setup a Window98 based bootdisk, try to make some net use
an then ould be able to have shared folder (from a pc running a OS from harddisk) there the Boinc client will resides.
But as on seti1, I think the crunhing time will be higher.

made in the past:
Windows 98er installation, reduced to under 100MB, could be run from a
"ramdisk" (virtual Harddisk, located in the main memory) this image
can be loaded also from an CD or DVD or from (with net use) from a other PC
some month ago, under seti1, a small pc was running quit well, booted from diskette and loaded image from CD, after unpacking, I had a "C:" from
the ramdisk, It was possible to make this with 256 mb RAM.

(on the other way: a realy naked windows (with less patched) will may be
turn into sleepless nights, because this client should be attached to the internet 24h/day also... security)

-2-
some one tried with knoppix & wine?
(Unix booted from a CD or DVD)
I belive time will come and some users will share there experiences.

Also interested on "silent" system

-3-
for the 5 mentioned "migrated systems" (w2k), it touk nearly more time to install all the provided patches als the initial installation..

due that now all clients are attached (by a proxy) to the internet, the security aspect increases (in my point of view, without getting panic)

happy crunching

Richard
ID: 6620 · Report as offensive
Jason Ebacher

Send message
Joined: 16 Aug 99
Posts: 19
Credit: 5,692
RAC: 0
United States
Message 6661 - Posted: 11 Jul 2004, 21:44:28 UTC
Last modified: 11 Jul 2004, 21:48:28 UTC

Bionic view sounds like a start... at least this will provide a central location to report to.... Now, the Linux Boinic client is one executable it has no dependancies on the OS just run it as the old version was... now if we could get a windows version of boinic in the same fashion we could just replace the old executable with the new and we would be set...

Any one good with compiling the code on the boinic page? if we could get a single stand alone version that worked like the old version in the sence that it kept all its files in the one directory we would be set...then with my enviroment (windows based BartPE farm) I could retain all the curent feature I have and gain a few more probably...

Any one here up to recompiling the bionic app for all us "farmers"...

If this can be done I will be able to add 5 to 10 more PIII class machines to my farm as they will all run from the BartPE cd and need only a nic mobo RAM and a cd drive... simple, automated and cost & power effiecient.

How would we be able to get the folks at SETI to make a version for us farmers? that works in a simular fasion to the old CLI client...

If any one wants any info on the BartPE way of setting up a farm let me know, I can't redistribute the OS as it is Microsoft stuff but it is very easy to setup and make your own bootable SETI OS cd's...at least with the old client its easy.

The reason I use this method is it's simple, automated and very fast, everything is in memeory, requires less parts to build a node, and its the windows version and is the fastest way to run WU's, Im getting 3.2 hours a WU on a 1700+ AMD with 128mb of RAM and a CD drive.

Thanks,
Jason
ID: 6661 · Report as offensive
ric
Volunteer tester
Avatar

Send message
Joined: 16 Jun 03
Posts: 482
Credit: 666,047
RAC: 0
Switzerland
Message 6702 - Posted: 11 Jul 2004, 23:13:16 UTC - in response to Message 6661.  

Hi Jason,

> Any one here up to recompiling the bionic app for all us "farmers"...

from an ealier job I'm used to build prebuild kits and installations
packages for MS environments.

Hmm.... was looking since month for a kind of solution with bootable cd.
Had one from CHIP.

so far only see solution with a reboot every 24 h.

Thanks a lot for the Tipp about BartPE, i'ts was new 4me, I 'm reading website
www.nu2.nu/pebuilde for more info.

But this will still be a Window98 environment, but booted from cd.
How are yours cunchtime under 98 versus other windows system on the same pc?
(on a mesured 1 Ghz pIII, same seti1 client, win98 apox. 3,3 WUs/day, w2k 3,8 WUs/day, data from SetiQ)

==> the idea for farmersbest bootable CDrom should be verified and tested

ID: 6702 · Report as offensive
Jason Ebacher

Send message
Joined: 16 Aug 99
Posts: 19
Credit: 5,692
RAC: 0
United States
Message 6707 - Posted: 11 Jul 2004, 23:43:12 UTC
Last modified: 11 Jul 2004, 23:50:33 UTC

Yes there is a solution to stop the 24 hour reboot, doesn't matter if the app is run from a drive share on your network...This also makes reporting easier as al the data is in one share on your network..one machine and share to lock down , if your worried about security.

About the reboot, The machine just reboots and fires up the seti app again and it starts where it left off... If a win98 botable CD is faster than The BartPE solution That would work also but 98 has downfalls and BartPE is probably more stable as its base is XP or Win2003... The entire ISO image that is made from the BartPE builder is 149mb so this is a very stripped down version of the NT OS, so it should be faster than a standard win2k install remember not even explorer is running on th BartPE install. just the NT core and the NU2menu is running...very clean.. I have it running on a Pentium 1 166mhz with only 96 mb ram and it runs fairly well, Imagine XP on a p166 that boots from cd and its usable..

Besides a boot cd, The only other solution I see for a easy to run and maintain farm is a beuwolf cluster and a specially compiled parallel processing BOINC app designed for beuwolf clusters... I have also looked into the Cluster Kenopix distro but this doesn't spread the one app ove multiple macines it just would allow you to run 5 apps on one pc and push the job to all the other nodes... But woul give you network boot on all the other nodes... Less hardware for this..

I and other "farmers" want a easy to monitor and run system.. idealy at the maximum speed we can making the best use of the resources consumed like hardware and power usage..

My farm now is 8 nodes..and 1 is on the boinc client (testing right now) everything else is on the old client, nost of these systems are caseless and SETI stack or shelf style, My goal is to find a better wat to run BOINC and add my 5-10 nodes on BOINc and then switch the rest of the remaining over to BOINC after the new nodes are up.. with the rebuild of the switch to BOINC and the additional boxes i will also be moving the farm to the garage, as it will be taking up to much space and adding tons of heat to the house.

Jason
ID: 6707 · Report as offensive
Jason Ebacher

Send message
Joined: 16 Aug 99
Posts: 19
Credit: 5,692
RAC: 0
United States
Message 6716 - Posted: 12 Jul 2004, 0:09:13 UTC
Last modified: 12 Jul 2004, 0:18:16 UTC

Actually on the seti farm boot disc idea I have a boot floppy based on the freesco router that is a floppy it boots and runs the old client... its linux and has a web interface but has limitations like, if the power goes down you loose everythng you have done in the current WU.. but it works..

I like the idea of this the best:

A Linux distro with the beuwolf software setup out of the box that automaticly configures the nodes via network boot and automaticly start the custom beuwolf version of he BOINC app upon boot.... this would allow you to install on one pc and all the nodes would work on that one process... any new nodes would be added to the cluster would just add to the power of the cluster.

Now Linux is probably the only way we could do this as it is freely distributable.. If we could provide a single disk image that a person would download, burn and boot, that would setup th master node of the cluster, and this master node would configure all the other slave nodes... this would make a standard platform for any one running a dedicated boinic farm... as known most of the top WU returners are running dedicated farms, and this would make it even easier... so easy in fact that any one can build a farm... and low cost also, the only thing you would need is one pc as the master that contained a harddisk and the nodes would be simply a Mother board with a bootable network card, very low cost and one central place to administer the whole cluser... Simple, cost effective and fairly easy...that would be the goal.

If you made a distro like this we would probably be making a huge impact on the seti project, If it easy to do your going to get alot more persons doing it. If you can get SETI's buy in and have them add for instance a paramiter in there database for "running SETE farm OS" to allow additionals downloads of WU to cache, this would allow you cluster to get say 10 WU at a time and when you only have left to get a additional 10... I guess if you could run WU's in 25 minutes, there would be good reason to support it on there end.

Any One else in for a Beuwolf build?

Jason
ID: 6716 · Report as offensive
Jason Ebacher

Send message
Joined: 16 Aug 99
Posts: 19
Credit: 5,692
RAC: 0
United States
Message 6719 - Posted: 12 Jul 2004, 0:19:56 UTC - in response to Message 6712.  

> > SO what are others that are dong farms,
>
> Dong farms? It must be quite interesting at harvest time!
> src="http://dapence.com/hosted/marvinhover81.gif">
>

Opps, Yes harvest must be very interesting, probably not as interesting as when we find ET.

Jason
ID: 6719 · Report as offensive
ric
Volunteer tester
Avatar

Send message
Joined: 16 Jun 03
Posts: 482
Credit: 666,047
RAC: 0
Switzerland
Message 6734 - Posted: 12 Jul 2004, 1:20:18 UTC

downloaded latest BartPE,
just made a bootable cd from (without an installed boincclient) and
stated it. O yea look like a better windows nt 4 Boot enviroment. Great.

Have still some problems to get into the local network, NT recognized the
nic, put the IP and subnet, changed the systemname,
should be on the local network, but can't ping a other pc nor can net use
to an other.

I have to spend a litte bit more of work, but an interesting way.

so far I did not read everything in the doku, is BartPE booting every 24h or not?

Did you build one cd for each system or a kind of master and then
added the Registery Settings to indiviualize/personalize the PCs?
what could be the best way?

thinking the bootable NT should only be able to boot, attaches a drive from network and run "what else needed"

ID: 6734 · Report as offensive
Jason Ebacher

Send message
Joined: 16 Aug 99
Posts: 19
Credit: 5,692
RAC: 0
United States
Message 6745 - Posted: 12 Jul 2004, 2:10:59 UTC
Last modified: 12 Jul 2004, 2:30:42 UTC

On BartPE, I set it up to pull DHCP, and the system will generate random names for the pc on boot...

Im using the old seti client on a mapped drive, so what I did was had it boot and config the network settings automaticly.. map a drive the "s" drive, then call the seti client from that mapped drive, just like you went to, run and typed "s:/seti.exe"

This runs the client "from the mapping" and then when the PC reboots at the 24hour mark you don't loose that WU because th client and associated files are on the server where the maped drive is... if you ran it localy you would loose the WU you were running because the appand files are in memory.

What I did for my other nodes was simply changed the call to the app on the mapped drive to "s:/seti2.exe" and "s:/seti3.exe" etc and just made a iso for each node and burned it to disk and booted it..

In the end I get a OS that boots its particular seti app in a seperate directory form the others on a central server share..all automaticly.. and then I use setiwatch to monitor my farms progress on a central pc.

the problem with BOINC is that th setup puts files in the windows directory and has to run from where it was installed, you could get it running on BartPE but any time you rebooted or lost power you would loose your WU work, as everything is in memeory... so if we could get a CLI of the new BOINC client that made no hooks to the OS like the old one... we could then just replace the OLD executable with the NEW executable and be off and running on BIONIC... The Linux client for BOINC is a single CLI that requires no install..why isnt there one like this for windows?

I also found some info on VPM (Virtial Parrallel Machine)... It works on Unixlike and Windows OS's and creates a parralel computing enviroment... havent messed with it and would require a full windows install on all clients (as far as I know from what I read) Not really anything like the Beowolf setu describe in a previos post here in this thread.

Yes By default the BartPE disk will reboot after 24, thus my automation of the dhcp and start of the SETI client... there is a patch you can find on the net however it is in the gray area of legality as you doing what M$ doesnt want you to do, a Windows XP bootable CD.. Because thats what it basicly is, I have seen full explorer, IE and almost all features of windows running from CD.

Jason
ID: 6745 · Report as offensive
Jason Ebacher

Send message
Joined: 16 Aug 99
Posts: 19
Credit: 5,692
RAC: 0
United States
Message 6755 - Posted: 12 Jul 2004, 2:49:27 UTC
Last modified: 12 Jul 2004, 2:50:33 UTC

Ok just for kicks I copied the Boinc Dir from my test BOINC machine to my network share... then on one of my nodes ran the BOINIC CLI.exe and it ran!

IS this OK to do will I suffer any "problems"

If this works I'm set.. I will be able to do what I want to do..YES!

Jason
ID: 6755 · Report as offensive
ric
Volunteer tester
Avatar

Send message
Joined: 16 Jun 03
Posts: 482
Credit: 666,047
RAC: 0
Switzerland
Message 6761 - Posted: 12 Jul 2004, 3:03:41 UTC - in response to Message 6745.  

Thanks for the infos.
It's like to be at LEGOLAND
> On BartPE, I setit up to pull DHCP, and the system will generate randome names
did also, static and DHCP, so far not on the network
==> in the doku found, additional plugin neeeded for some networks function,
"tcp/ip support" this could also be a hit to get the system on the net.

> the problem with BOINC is that the setup puts files in the windows directory
> and has to run from where it was installed, you could get it running on
not also a realy a unsolve-able problem. any kind of application (looking deeper, we got at least 2 of them, the "Boincclient" and "that" was the attached project brings in )) addes/makes files, subdirs and registery setting.
We have to find out what is going on while the installation and while the client is running. It's a work but can be done.
found in the registery:
(my installation is from "F:")

[HKEY_CURRENT_USER\Software\BOINC]
"ClientPath"="F:\\BOINC\\boinc_gui.exe"
"ClientDir"="F:\\BOINC"
"WinDir"="F:\\WINDOWS\"

and
[HKEY_LOCAL_MACHINE\SOFTWARE\Microsoft\Windows\CurrentVersion\App Paths\boinc_gui.exe]
"Path"="F:\\BOINC"
@="F:\\BOINC\\boinc_gui.exe"

and also found infos about 2 shared DLLs (and registered)
(maybe also in the run section)

when the BartPE "OS" reboots, it will install the needes files in the windows dir (or better into the ramdisk) and added a path to there or they stay on a network, as I understood the client will update itself??

> Yes By default the BartPE disk will reboot after 24,
for me that is not realy the problem, if they do ot themselfe ok. Are we shoure
the the boincclient well restart at the last checkpoint?

grey can be a great color like others else. interested when I'am ready for the last .iso :-))

Reading the infos about creating plugins and
ID: 6761 · Report as offensive
Jason Ebacher

Send message
Joined: 16 Aug 99
Posts: 19
Credit: 5,692
RAC: 0
United States
Message 6765 - Posted: 12 Jul 2004, 3:14:17 UTC

My BartPE disk surfed the web and the remote destop client worked first time after I setup my network atributes, ny first mod to the stock ISO was automating the DHCP request, its very simple to do, next was adding the mapping to my data share where the seti clients were located.. then it was simply getting the app to start automaticly from the share... thats it and it works great.

Now after my discovery of the CLI not needing dlls ect from the os im trying to build a version that calls the BOINC CLI and we'll see if that works.

Jason
ID: 6765 · Report as offensive
Profile MAOJC

Send message
Joined: 31 Jan 00
Posts: 11
Credit: 991,339
RAC: 0
United States
Message 6816 - Posted: 12 Jul 2004, 6:02:39 UTC
Last modified: 12 Jul 2004, 6:06:13 UTC

I have a farm of 26 machines based on LTSP (ltsp.org). They boot from the network with a combination of PXE and CF flash etherboot (rom-o-matic.net) images that mount a linux server via TFTP. All I/O is NFS and all data is stored in a single directory on the server with sub diretories by machine name. The server is a SDRAM XP1800 palimino with a 60gig disk with DHCP, DNS and TFTP daemons. Server load averages hover around 0.30. It works very well and I did run seti classic under wine on a similar configuration. I did standardize on a single NIC, realtec 8139, to simplify the installations. They all run 2.4.20-8 kernel.

The farm is mostly AMD XP2500 @ 200 (XP3200+) with a few XP2400's included. The boards are mostly a combination of Soltek NV400 and Abit NF7. The Abit are nice as they will boot headless no problem. The rest of used $5 PCI VGA cards. My total cost per client is at about $200 but I have one XP2700 on an ECS KT600 I got from Fry's for $99. My only limitation to expansion at this point is the electic service to this room. I have added a 15,000 BTU AC unit to the room to help cool this crazyness. I may consider a large evap water cooling setup for next spring to move the heat outside.

I wrote some simple scripts to start boinc at boot, and monitor the progress from the server while making sure the client platforms are healthy.

Probably my next steps are to get an optimized seti client running for improved performance and write a web based system monitoring and management tool.
ID: 6816 · Report as offensive
Jason Ebacher

Send message
Joined: 16 Aug 99
Posts: 19
Credit: 5,692
RAC: 0
United States
Message 6839 - Posted: 12 Jul 2004, 7:18:57 UTC
Last modified: 12 Jul 2004, 7:31:15 UTC

Wow, now thats a farm.. Il get there in a year or so...

I have discovered that the BOINC windows CLI will run from a maped share on a central server so im making boot cd's based on BartPE and running from my server... This shoudl work well in combination with BOINC View..

What I couldnt believe is that my BartPE based machines are really crunching quite fast in comparison to non PE version on the same hardware... That slim OS really helps.


<a> [/url]
ID: 6839 · Report as offensive
Jason Ebacher

Send message
Joined: 16 Aug 99
Posts: 19
Credit: 5,692
RAC: 0
United States
Message 6849 - Posted: 12 Jul 2004, 8:16:29 UTC

Ok I thought I had it but the Bart PE version will not run the client, I think it has to do with running everything in memory.. But a normal install of windows will work.. So i do have options... I will keep working on the BartPE version... Maybe installing it to hardisk will work.. its only 140 mb so any old disk should work just fine.

Jason
<a> [/url]
ID: 6849 · Report as offensive
ric
Volunteer tester
Avatar

Send message
Joined: 16 Jun 03
Posts: 482
Credit: 666,047
RAC: 0
Switzerland
Message 7187 - Posted: 12 Jul 2004, 22:25:37 UTC

hello to all,

Already alive, but not 24 H in front of PCs.. sometime I used to sleep

could solve the PE Network access problem.
On this PC there are 2NICs, MB build in and an aditional realtek for dsl or more
It take a time to put the right cable to the right place..

Could get all nessesary files for the network support

Situation in short:
the test pc, booted from PE is running now with the boincclient
(from b: as the ramdisk), copied installed boincclient from from a shared folder, from an XP PC ) made shure needet dlls are in the boincclient directory,

added 2 regfiles ( took from a normal running installation,changed path intp b:
the started the boinc_cli.exe, defined the proxy, stopped /started the b client
and the attachend to project seti
the client got berkeley, talkes to him, shoes me credit and now
waiting for a wu, "not work from project..."

For going so far, I had to made several steps by hand.
There is work to do, but think can be automated.

The benchmark shows a to high result for HT, 1941 /4192 on a 3GHz
I think there is no HT support for PE

(only for a test, AMDs are waiting)

on the coumputer page
69603 minint-o2rxth 0.00 0.00 Intel(R) Pentium(R) 4 CPU 3.00GHz Pentium Microsoft Windows XP Professional Edition, (05.01.2600.00)


I want to give a large tankyou to all who gaves infos and spend time to type in experiences, that is exactly the target of a board...

you have the luck to be familiary with pe, that helps a lot

still no work from project

On the CD I added a first Try with the boincclient (naked) with a file commander, the iso indreases to about 172 MB, there is plenty space left

greeting and happy crunshing (whatever your OS is named)

Richard
ID: 7187 · Report as offensive
Jason Ebacher

Send message
Joined: 16 Aug 99
Posts: 19
Credit: 5,692
RAC: 0
United States
Message 7216 - Posted: 13 Jul 2004, 0:58:46 UTC
Last modified: 13 Jul 2004, 1:23:28 UTC

My issue was when I would get the client running it would download the new WU then it would error out like it was needing somthing from the OS to start the processing... So I tried running from the file share, the exact same app and directory, this time on a full fledged windows OS, and there error never happens.. it ran for about 12 hours today while I was at work.

You stated:

"added 2 regfiles ( took from a normal running installation,changed path intp b:" What were these files, where are they located etc... this may be what im missing from my install...

The only time it errors is when it first starts crunching the WU, after downloading it.

As for Hyper Threading, there should be support for HT in BArtPE as the kernal is the same.. but like the regulat instal of BOINC you have to run two CLI's to use the other virtual CPU..

Jason



<a> [/url]
ID: 7216 · Report as offensive
1 · 2 · Next

Message boards : Number crunching : Rebuilding the farm


 
©2024 University of California
 
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.