Source-Changes archive

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Old Index]

Re: CVS commit: src



On Sun, Dec 21, 2003 at 09:43:11PM +0000, David Laight wrote:
  | >   This is up to 30% faster when populating an empty /dev,
  | >   and nearly 2x faster when re-running on an existing /dev.
  | 
  | Did you benchmark MAKEDEV -s | pax -M (or | mtree) ?

Oh Yes.

Here's the numbers on a P4-2400;
the first time is populating an empty directory,
the second time is rerunning MAKEDEV in that directory.

1.  old  MAKEDEV all
                1.516u 1.522s 0:03.18 95.2%     0+0k 28+4363io 0pf+0w
                1.625u 1.700s 0:03.36 98.8%     0+0k 0+6116io 0pf+0w

2.  new  MAKEDEV all
                1.184u 1.309s 0:02.50 99.2%     0+0k 28+3799io 0pf+0w
                1.032u 1.116s 0:02.09 102.3%    0+0k 0+39io 0pf+0w

3.  new  MAKEDEV -s all | pax -w -M | pax -r -pe
                0.189u 0.316s 0:00.51 96.0%     0+0k 0+3767io 0pf+0w
                0.164u 0.472s 0:00.66 95.4%     0+0k 0+5559io 0pf+0w

4.  new  MAKEDEV -s all | mtree -e -u
                0.040u 0.160s 0:00.21 95.2%     0+0k 0+3769io 0pf+0w
                0.061u 0.010s 0:00.07 100.0%    0+0k 0+1io 0pf+0w


The problem with 3 & 4 is that they lose the benefit of the "mknod -r",
which is to retain permissions and ownership of existing devices.
That said, we could use pax|pax or mtree to populate an empty /dev
in sysinst and possibly the "init mfs /dev" trick.

Cheers,
Luke.

Attachment: pgpJ5kTIN5GT_.pgp
Description: PGP signature



Home | Main Index | Thread Index | Old Index