From: pdh@netcom.com (P D H) Subject: Re: New feature for the filesystems. What do you think ? Date: Mon, 19 Apr 1993 06:18:20 GMT
rajat@watson.ibm.com (Rajat Datta) writes:
>Let's assume we only have 2 blocksizes. One is the pagesize, another
>is 1/2 the pagesize for compressed blocks. When the filesystem issues
>a write request, the device driver compresses the block and chooses
>one of the two blocksizes. It writes out the block to one of two
>different regions. Obviously, the device driver has to be capable of
>changing the block number that the filesystem thinks it's writing out
>to, which is a change in the device driver interface.
This assumes all blocks can be compressed to 1/2 size. I had hoped
such assumptions were readable only in the Nation Enquirer, but I guess
I am wrong.
>Anyway, I guess compression is going to be built in into the disk
>hardware pretty soon. IBM's high-end disks have started doing just
>this. Since the disk controllers have a fairly large amount of cache
>(to let slow-moving media keep up with the rapid processors), there is
>no performance impact if the data is compressed as it's written out.
It's one thing compress a block of data. It's yet another thing to take
advantage of that compression by writing something else in the rest of
the block. And since you don't always get 512 bytes from 1024 bytes of
data, the remaining size available cannot be guaranteed to be 512 bytes.
Assume the hard disk has 1048576 blocks of 1024 bytes each. How many
blocks do you make the file system for?
-- | Phil Howard, pdh@netcom.com, KA9WGN Spell protection? "1(911)A1" | | Right wing conservative capitalists are out to separate you from your MONEY | | Left wing liberal do gooders are out to separate you from EVERYTHING ELSE!! | +-----------------------------------------------------------------------------+