URL:
<http://savannah.gnu.org/bugs/?func=detailitem&item_id=17193>
Summary: performance of `dd' w.r.t. different block sizes
Project: The GNU Hurd
Submitted by: tschwinge
Submitted on: Monday 07/24/06 at 16:44
Category: None
Severity: 3 - Normal
Priority: 5 - Normal
Item Group: None
Status: None
Privacy: Public
Assigned to: None
Originator Name:
Originator Email:
Open/Closed: Open
Reproducibility: None
Size (loc): None
Planned Release: None
Effort: 0.00
_______________________________________________________
Details:
20:03 <antrik> with "dd if=/dev/hd0s6 of=/dev/hd0s8 bs=512" I get the same 20
kbyte/s as with cat
20:03 <antrik> but with bs=1k, I already get 75 k...
20:03 <antrik> 130 with 2k
20:04 <antrik> 230 with 4k
20:04 <antrik> and so on
20:04 <ness> Does copying the files stress ipc (I'd guess)?
20:05 <ness> Because overhead is smaller for larger payload sizes...
20:08 <antrik> well, with larger block sizes, the disk seeks considerably
less; looks like each single copied block is written immediately...
20:09 <antrik> that doesn't explain the unexpected jump between 512 byte and
1024 byte block size, though
20:10 <antrik> maybe with 1024 it is fast enought so the disk access is
optimized a bit at least, or something
20:11 <antrik> or it's some sheduling issue
20:27 <antrik> also, with sizes <512 there is no further slowdown
20:27 <antrik> (well, except when going <16...)
23:21 <antrik> interesting... if I give a larger blocksize to dd, the
relative CPU consumption rises compared to that one of pflocal...
23:21 <antrik> and the CPU consumption of /dev/zero drops nearly to 0
_______________________________________________________
Reply to this item at:
<http://savannah.gnu.org/bugs/?func=detailitem&item_id=17193>
_______________________________________________
Message sent via/by Savannah
http://savannah.gnu.org/
_______________________________________________
Bug-hurd mailing list
Bug-hurd@gnu.org
http://lists.gnu.org/mailman/listinfo/bug-hurd
