Re: Fragments in Ext2 fs

Linus Torvalds (Linus.Torvalds@cs.Helsinki.FI)
Mon, 10 Jul 1995 07:58:26 +0300


Brian Haney: "Fragments in Ext2 fs" (Jul 7, 8:07):
> What is the status of support for block fragments in the
> Ext2 filesystem.
>
> Has it happened and I missed it?
> Is it a did issue?
> Does anybody care? (I sure do! It eats up a helluva lot of disk!)

As far as I'm concerened, I'd rather not see fragments. They aren't
worth it, IMHO.

- They add unnecessary complexity
- the default linux blocksize is just 1kB, which is just about the
default fragment-size for BSD systems, I do believe (8kB blocks, 1kB
fragments?)
- througput through larger blocksizes can be just about offset by using
scatter-gather device drivers (and linux has them) and some extra CPU
overhead.

(If you didn't guess, I think the CPU overhead isn't that bad. Tastes
differ. I prefer a clean approach without the BSD baggage, and
according to benchmarks linux is faster than BSD on most file operations
anyway: if you want to optimize, I think stdio is more optimizable).

Linus