Re: Most efficient way to write a large matrix to disk
- From: "Lorents" <fdsa@xxxxx>
- Date: Tue, 26 Aug 2008 20:18:10 +0100
glen herrmannsfeldt wrote:
[...]
For reasonable N, I would not expect a large difference.
Also, some systems might want a buffer big enough for a single
WRITE, which would give CBC a big advantage...
(snip)
Ok, good to know :-)
Yes, it sounds like you are filling the disk cache with
the first case, and then for the second it actually waits
until it is written to disk.
I did a few more tests and yes, the (software, I presume) disk
cache / write behind makes my rudimentary benchmark strategy
almost useless...
For very big runs clocked times increase a lot probably because
all the memory is used to store the matrix and there is no memory
left for the disk cache.
Most of the work is in library routines that don't change
with optimization level. Disk drive speeds don't change,
either.
Yes, I thought as much too.
Thank you,
Lorenzo
.
- References:
- Re: Most efficient way to write a large matrix to disk
- From: glen herrmannsfeldt
- Re: Most efficient way to write a large matrix to disk
- Prev by Date: Re: busting sp datatypes
- Next by Date: Re: Most efficient way to write a large matrix to disk
- Previous by thread: Re: Most efficient way to write a large matrix to disk
- Next by thread: Re: Most efficient way to write a large matrix to disk
- Index(es):
Relevant Pages
|
