[Users] Probable speedup of simulations - Ram File system

[Frank Loeffler]

Hi David,

Thanks for your suggestion. In short: I wouldn't expect anyone here
having tried this, but I believe it would work. Assuming the data fits
into ram that is. Often that "if" is too big.

A longer answer:

On Mon, Dec 24, 2012 at 01:52:47PM -0600, David N. Bradley wrote:
> After compiling the cactus and ET, I ran the static_tov simulation in
> a ram file system.  It seems to run faster, since it did not have to
> spin up the harddrive for each iteration and append the data.

Does your hard drive really spin up and down between data writes? Well -
in case it really does I imagine that yes, in that case a ram file
system could help in some cases.

However, I have to say that most, if not almost all Cactus users run
on machines with disks that never sleep. On production clusters the
file system you are writing to is a network file system most of the time
anyway (again, with either disks that spin all the time anyway or SSDs
as buffer).

Now on the other extreme you have laptop systems. Running a simulation
on battery for a long time is a bad idea anyway, so I assume you are
plugged in. Especially then I wouldn't let the disk spin down at all as
long as you are working on it. It just creates lags and wears the disk
more than letting it spin idle.  Personally I don't let it spin down
even on battery, but that is personal preference (When on battery I am
on the machine, working. When I work I do write to disk quite often.
Spinning a disk up needs time and power - so I don't bother letting it
spin down in the first place.)

Frank

-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 836 bytes
Desc: Digital signature
Url : http://lists.einsteintoolkit.org/pipermail/users/attachments/20121225/1660d9ab/attachment.bin