New Tool

[Hans Arne Iversen]

A new tool is out
KDFold for linux and windows
(The linux verson isn't ready yet)
Can monitor up to 100 clents
small memory footprint
can monitor genome and folding
Generate stats in html format

You can read about it here
http://www.hardforum.com/showthread.php?s=&threadid=197119&perpage=20&pagenumber=1

and download it from here
http://www.plasticstudios.org/robsdelphipage.html

[wbierman]

I've tried it works great on Windows!

[Jodie]

Guess I need to samba and nfs my machines.. There goes some cycles...:D

Good catch! Maybe I'll see about porting it into my project and saving some time... But Delphi though?

[pelligrini]

Looks very cool.
Gonna see how it works. :)

[Dave S]

cheers man gonna give it a go:cool:

[pelligrini]

Works farily well
Doesn't like large font settings in windows though

[wbierman]

But it does work! At least across my network and even shows when a machine is down. I like it!

[ohms18k]

I like it!

Me too wbierman :)

[Jodie]

Now I'm intrigued - I'm installing samba on all the linux machines now...

[wbierman]

Here is a pic of my farm via KDview

http://63.204.18.114/sig_test/pics.htm

[The GRGI Borg]

Thank you very much for this beautifull utility....

all my pc say to you thank you and my legs too.. :)

Good job guy.

Marzio

[X-Calibur]

JFYI:

you can customize the logo by saving a pic w/ the name logo.jpg the the folder where KDFold is stored.

I used our logo, it's not too bad, a bit compressed.

No pic, as Photoshop is not reinstalled....

CHeers !

X

[The GRGI Borg]

i'll start in 10 minutes prepare a good picture.... i hope.

bye

Marzio

[Jodie]

WooHoo! From the "for what it's worth department" --

I *finally* got enough back into my kernel to allow me to samba mount a linux machine. Exactly how I got it working, I'm uncertain - think I'll tar up all of /etc and move it straight across to the other machines. I have seven of, umm, 18 or 19 processors into kdfold now... Only two of which are linux machines atm, but all the rest of the ones to add are...

Here's a preliminary screen shot -

http://www.interactdevices.com/tgc/llair.jpg

(Link only as it's still 80Kbytes. I didn't do anything to 'smallify it')

Cool Tool! I have a couple of suggestions to send him though...

[Bruce]

Originally posted by Jodie
WooHoo! From the "for what it's worth department" --

Cool Tool! I have a couple of suggestions to send him though...

Looking good.

Ask him what's that stuff about msgs.

[wbierman]

I see you have a newer version...

[Jodie]

Of course, I do, Will! Would you expect less? :p :cool: :D :D :D

[wbierman]

Never

[pelligrini]

Has anyone installed a newer version over an older one?
I'd hate to have to enter in 44 cpu's again.

BTW have you seen this one (http://130.94.169.148/vbulletin/showthread.php3?threadid=1671) yet?

[Jodie]

I've been waiting for the release, even posted to their board in german, asking for it. [grin] (as well as translating a post or two for others)

I've got a kinda nifty hack running right now - the obvious W2k running the kdfold client, with the linux machines on Samba. What's spiffy is that I have an old copy of PCAnywhere for CE (may be discontinued, not sure) running over the wireless modem to the W2k box. I can see the kdfold client (sloooowly at 14.4kbit) and if any of the machines are down, run telnet across the wireless to them and tar up the bad wu and -clear. It's suboptimal, but it's a marvelous stop-gap!

I'm back to adding more samba shares! If I can get all the machines into kdfold and handle them remotely, I'll feel good about adding hardware again...

[wbierman]

Yes I have seen Genome Spy. I'm going to load it tonight and try it.

[pelligrini]

Jodie- I had my first v99 gene crash on my main machine yesterday, but I restarted the client and it finished normally. Do you automatically -clear yours?

[Jodie]

yup. I don't want to waste the 'filter' time... :p :D

I tar them up, move them to a shared directory on windows, and let windows finish 'em off... Atleast until Stephan fixes that darned linux bug...

BTW, if you save your .psx and .pwx files, you should be pretty safe in the upgrade. They're raw text, human readable and easy to figure out what they are... If something has changed, it'd be easier to change the file...

If you want to send me one of each, I can compare them to mine and see if anything has changed...

[Jodie]

Originally posted by pelligrini
Has anyone installed a newer version over an older one?
I'd hate to have to enter in 44 cpu's again.

BTW have you seen this one (http://130.94.169.148/vbulletin/showthread.php3?threadid=1671) yet?

So far, I'm pretty unimpressed with GenomeSpy. Seems to crash every time I try to mount a network share in it, for me... The UI is great, though... Tooo bad... Hopefully it will improve. I'm going to go back to adding machines to kdfold... :(

[pelligrini]

Originally posted by Jodie
BTW, if you save your .psx and .pwx files, you should be pretty safe in the upgrade. They're raw text, human readable and easy to figure out what they are... If something has changed, it'd be easier to change the file...
That's ok, I'm just being lazy. Been hanging around the kids too much. ;)

It wouldn't be that hard to just make a copy of my current installation.

[Jodie]

That too.:cool:

[pelligrini]

The latest version installs right over my earlier version keeping all the entered cpu's.

[ohms18k]

Possible Warning :(

If you installed KDview on a machine with the client running check out its time, the box i put it on, the client slowed drastically. Haven’t checked what the problem could be just shut KDview off. I hope it was just the box I put it on, running W2k pro.

[Jodie]

On a two processor machine running win2k and tracking 25 clients, I saw 47hrs of CPU time for one client and 39hrs for another. That bothered me, so I looked further. KDFold had used 8 HOURS of CPU time. Unacceptable - not at all what was advertised. GenomeSpy is worse yet...

I've nuked them and written a shell-script that basically does a hostname >foo.asc ; ps -ax |grep "ghclient" >> foo.asc |unattended ftp to a server. The server runs a cron every 10mins to cat those files together into one big file and cat an html header/footer onto the files.

You can then see how long the client has been up and whether or not it is. So far I've seen six cpu seconds in eighteen hours... More reliable measure, too... Also can be viewed from anywhere with a web connection. Simple and super-effective...

[dnar]

Pretty much what I do too, too easy.

[ohms18k]

On a two processor machine running win2k and tracking 25 clients, I saw 47hrs of CPU time for one client and 39hrs for another. That bothered me, so I looked further. KDFold had used 8 HOURS of CPU time. Unacceptable - not at all what was advertised.

Good to see you are on top of things Jodie :cool:

Would love to see that script you wrote.

[dnar]

Originally posted by ohms18k




Would love to see that script you wrote.

Yeh!, share and share alike :D

[Jodie]

Sure! I'll go push it up to the web site here and post a link. It's really in two parts and is just a dirt-simple shell script... Server-side and Client-side.

You could also run it under windows using Cygnus unix-tools port...

[Jodie]

Real quick, before I do so, have a look at this... I restarted kdfold and let it run for a bit - look at the comparative utilization - stealing about 19% of one of the CPUs! And these are 1Ghz PIII's with 1.25G of physical ram! (22 clients monitored.)

http://www.interactdevices.com/tgc/cpu-usage.jpg

[eldiablo]

Originally posted by Jodie


So far, I'm pretty unimpressed with GenomeSpy. Seems to crash every time I try to mount a network share in it, for me... The UI is great, though... Tooo bad... Hopefully it will improve. I'm going to go back to adding machines to kdfold... :(
what exactly do you mean by "mount a network share"? do you mean monitor a mapped genome drive/directory on another machine on your network? this is what i am doing and i haven't had any problems.

On a two processor machine running win2k and tracking 25 clients, I saw 47hrs of CPU time for one client and 39hrs for another. That bothered me, so I looked further. KDFold had used 8 HOURS of CPU time. Unacceptable - not at all what was advertised. GenomeSpy is worse yet...
i wouldn't exactly call genome spy efficient, but it sounds like it is much worse for you than it is for me. what interval did you set to monitor your clients (not the "refresh" section in "options", the "client refresh rate" under the "edit" function.)? with 25 clients, you would probably see a large improvement by backing down the refresh rate.

[Jodie]

Originally posted by eldiablo

what exactly do you mean by "mount a network share"? do you mean monitor a mapped genome drive/directory on another machine on your network? this is what i am doing and i haven't had any problems.



Yup. I mount 22 drives on Linux machines with drive letters shared via samba. (sorry, "mount a network share" dates me, goes back to my time at Novell. . .)

I then assign one in GenomeSpy. It goes off to think and never returns. Doing the same works in KDFold... I get the same results by not assigning a drive letter, but rather using absolute path (\\foobar\genome) which also works on KDFold but goes away and never returns on my Win2k machine running the 'spy client.



i wouldn't exactly call genome spy efficient, but it sounds like it is much worse for you than it is for me. what interval did you set to monitor your clients (not the "refresh" section in "options", the "client refresh rate" under the "edit" function.)? with 25 clients, you would probably see a large improvement by backing down the refresh rate.

I'd need to go back and look at spy again to answer that question. Believe I left everything at default and tried to map a remote drive... I added my two local clients (dual proc) fine, but as soon as I add the first remote, POOF, it goes to lalaland and 10mins later still hasn't returned.

As for KDFold - I rained on its parade! I gave the genome clients an affinity to 0 and 1 respectively, bumped their priority to 'Above normal' and put KDFold on proc 0/1 with a priority of 'below normal' NOW it's being a bit more reasonable... Still updates fine, as far as I can see... It's used 15 seconds of CPU in 4hrs.

It takes about 5 seconds to switch screens on it, and any lower priority and its graphics never update... I'll try this until I get back home and see how it goes...

Thank you kindly for the reply and your input!!!

[Bruce]

I had a similar experience with kdfold -- well, sort of similar. I put it on a really slow and small machine and tried to monitor 10 machines. ghclient was virtually STOPPED.

My first assumption was that it used too much ram so ghclient was thrashing, trying to get paged back in. I though about asking what memory footprint it was taking in your environment but shut it down instead. (I'm also curious if it allocates noticeably more per machine.)

I'll probably try it on a machine with lots of ram.

There was a comment during beta worrying about how much processing it takes to get stateful task switching.

[eldiablo]

this is pre-caffeine so bear with me... ok, we're both running gspy on w2k. you are mapping to drives on *nix boxes using samba. i've heard of samba, but i'm not a *nix geek so i'm not familiar with it. perhaps that is where the problem lies. have you tried mapping a drive to another windows box and seeing if at least that works for you? remember, this is the first (ok, actually second - there was a quick fix for english language users) release of a free program, so try to not be too critical of it. :) i think it's actually pretty nice and just needs a little tweaking here and there...

[Jodie]

Originally posted by Bruce
I had a similar experience with kdfold -- well, sort of similar. I put it on a really slow and small machine and tried to monitor 10 machines. ghclient was virtually STOPPED.

My first assumption was that it used too much ram so ghclient was thrashing, trying to get paged back in. I though about asking what memory footprint it was taking in your environment but shut it down instead. (I'm also curious if it allocates noticeably more per machine.)

I'll probably try it on a machine with lots of ram.

There was a comment during beta worrying about how much processing it takes to get stateful task switching.

About 2.5M of RAM total. Not excessive at all...

[Bruce]

Originally posted by Jodie
About 2.5M of RAM total. Not excessive at all...

Then that wasn't the problem. I'll reserve judgement until there are more facts.