>Not really. The fastest systems today can't saturate a DVD, 100M
>network, or hard disk with bzip2.

Indeed.  (To quickly populate workspaces I keep a weekly tar.gz of
opensolaris; a fast system can gunzip this at 10MB/s easily when
unzipping to /tmp; +/- 12s of CPU to uncompress a 230MB file to approx 5x
as much data; bunzip2 would take minutes and would cause both
net bandwidth and disk bandwidth to be underutilized.

>I would happily exchange the 5% loss in compression for the 10x
>performance win.

Indeed.

>The cache only works for reads; logging only for metadata.
>
>(Casper said it's "transactionally safe". I would think that the
>relevant transactional unit here is the whole install.)

For upgrades the transaction unit is the package as upgrades can be
restarted.

However, what I meant was that the package data must be commited
to disk before the pkgadd can be considered done.

>That's not how I understood it. I thought the idea was that the package
>would be stored as a single file in datastream format rather than
>unpacked in filesystem format. Which would cut down a lot of the small I/O
>operations that are currently necessary.

Quite.

Perhaps it's time to test some of this.

Casper
_______________________________________________
opensolaris-discuss mailing list
[email protected]

Reply via email to