home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.unix.questions
- Path: sparky!uunet!decwrl!pa.dec.com!nntpd2.cxo.dec.com!nabeth!alan
- From: alan@nabeth.enet.dec.com (Alan Rollow - Alan's Home for Wayward Tumbleweeds.)
- Subject: Re: Berkely Fast File System
- Message-ID: <1992Sep12.003806.20983@nntpd2.cxo.dec.com>
- Lines: 44
- Sender: alan@nabeth (Alan Rollow - Alan's Home for Wayward Tumbleweeds.)
- Reply-To: alan@nabeth.enet.dec.com (Alan Rollow - Alan's Home for Wayward Tumbleweeds.)
- Organization: Digital Equipment Corporation
- References: <1992Sep7.085336.22530@nntp.uoregon.edu> <1992Sep9.025316.19275@nntpd2.cxo.dec.com> <BuBvL6.2MJ@gumby.ocs.com>
- Date: Sat, 12 Sep 1992 00:38:06 GMT
-
-
- In article <BuBvL6.2MJ@gumby.ocs.com>, obi@gumby.ocs.com (Kuryan "Obi" Thomas) writes:
- >
- >In article <1992Sep9.025316.19275@nntpd2.cxo.dec.com>,
- >Could Alan or
- >someone else discuss if any of this thinking is affected by SCSI disk
- >drives and if so how? My understanding of SCSI drives is that they hide
- >specifics like how many platters there are and so on. What's the strategy
- >used for dealing with these "block-stream" devices?
-
- My first cut at this showed me that my knowledge of disk driver
- internals is sorely lacking. My belief is that detailed knowledge
- of the geometry of disk is much less important than it used to be.
- Assuming you can get the underlying device to tell you the truth about
- the geometry. Certainly with the advent of read-ahead caches (and
- write behind if you're that trusting) in drives read performance
- is much less tied to geometry. Does it really matter if you have
- your transfers starting on track boundries, if the driver is reading
- four or eight tracks ahead of you?
-
- The basic principle guiding the organization by cylinder group
- is to keep the data close together. As long as the drive isn't
- wildly lying about it's geometry, then it still good to keep the
- data logically close together. That file system and partition
- boundries don't closely match up with supposed physical geometries
- probably doesn't matter much, use the geometry that will make
- best use of the available space.
-
- One that has been made partially obsolute is the work done to
- find the rotationally optimal location for a block. On drives
- with a read-ahead cache you don't block the spread out. You
- want them close together so the cache holds more of the data
- in which you are interested. If you do lots of sequential
- writting then playing with the rotational latency may be worth
- while, but you may hurt read performance later.
-
- Recent improvements in the implementation of the Fast File System
- by Sun and by DEC in ULTRIX V4.3 make this much less of a problem.
- That will be the topic of my next post in this thread. For now
- I'm off to see "Unforgiven"...
- >
- --
- Alan Rollow alan@nabeth.cxo.dec.com
-
-