PDA

View Full Version : Ramdrives and more speed for your folds.



Fozzie
08-24-2003, 04:35 PM
Just done a little test on my machine at home with the standard MS RamDrive.sys for XP.

I don't know if you will find the numbers significant but over the long haul for the guys who upload directly 24/7 it may prove interesting.

I ran the foldtrajlite -bench on the same system with the same generation one from the hard disk one from the ramdrive.

Hard Disk

Summary
-------
Usr time Sys time
-------- --------
Maketrj 10.391 0.344
Foldtraj 65.906 11.125

Ram drive

Summary
-------
Usr time Sys time
-------- --------
Maketrj 10.188 0.500
Foldtraj 64.609 11.672

Difference

Maketrj -0.203 +0.156
Foldtraj -1.297 +0.547


Overall it looks like a gain and for no money. :thumbs:

DocWardo
08-24-2003, 04:38 PM
Do switches like -rt affect the benchmark? and if so what switchs did you use?

Fozzie
08-24-2003, 04:42 PM
just the -bench command in the readme.

I am running it now from the RAMDrive with the standard switches on.

It seems to be using the same amount of RAM as it normally does though.


http://support.microsoft.com/default.aspx?scid=kb;en-us;Q257405&SD=MSKB&

Chinasaur
08-24-2003, 05:01 PM
RAM drive berry berrry dangerous without UPS....

Although...in YOUR case....

Use it and enjoy... ;)

DocWardo
08-24-2003, 06:42 PM
right but my thought was.. is there acutally a performance boost using the ramdrive, over using say -rt switch and setting -g 0.

it's gotta be using more ram (for the ram drive)
and what happens if you have to start caching data, will the ram drive fill?

Fozzie
08-25-2003, 04:11 AM
Your concern makes me feel all tingley :D

Doc I was only thinking of this process for those who have an always on Net connection.

The Ramdrive from M$ is only 32mb and as such for those who cache gens its no good.

However for those who upload as soon as the gen is finished then the speed advantages are there.

I was using the same switches with both benchmarks so there is an increase in speed wit the ramdrive.

Not much I'll grant you but if you have the RAM to spare (the faster the better) and as Chinasaur suggests a UPS then it could gain 1-2% more folds maybe moe over the long run.

GHOST
08-25-2003, 04:49 AM
Do switches like -rt affect the benchmark? and if so what switchs did you use?

i do not think switches affect benchmark. when i was sampling bench marks i watched my task manager and mem usage was 14000 or 20000. not 60,000 like when client is running.

DocWardo
08-25-2003, 08:28 AM
okay, but we were ALL caching generations last week when the server was down. we all cache when the server changes over.

I'm still not quite sure why this would be any faster than say running with extra ram which puts the entire project into ram to run. and with -g 0 which turns off progress reporting. so the only thing it will write is the data files at the end of the genreation for upload. is it really worth all of that work setting it up and possilbilty of loosing every drop of work when you have to reset for some reason?

Don't get me wrong this would probably be the way to go with a diskless linux node so that you don't have the slow link of an NFS read/write. but for the normal user running this on his home 'puter? a 1% gain will only get me 5400 points per day which is what 10 extra gen's per day roughly. it may run it faster, but don't forget to weight hte setback to the project should you loose a lot of data as well.

I'm not sayin it's a bad idea, just make sure that if anyone is going to run it this way they take precacations and know tthey run the risk of loosin it all. oh and I don't have a link but most ram drive systems have a periodic backup to disk funtion. that would be needed to be active for times like when the client updates. if it didn't back up to disk then the next time you started the ram drive you would have the old client.