1 |
Neat, although it would be more practical to simply use your |
2 |
usual RAM, well, pretty much completely analogously to the |
3 |
processor's L2 cache. |
4 |
|
5 |
A software solution rather than a hardware one. It probably |
6 |
wouldn't be generally applicable as the filesystem itself would |
7 |
have to support or in fact just implement it, (or the io layer |
8 |
understand specific filesystems, which is just troublesome) |
9 |
but it would be a workable improvement, or just extension to |
10 |
modern filesystems. |
11 |
|
12 |
|
13 |
> > Hrm... I've been wondering how much of a hard drive relief |
14 |
> > it would be to put the metadata of the top x most accessed |
15 |
> > files or just up to y MB in of it ramdisk. Considering hard |
16 |
> > drive flutter is a major bummer to performance these days, |
17 |
> > doing a flutter for common filesystem *meta*data seems |
18 |
> > rather silly - and a /lot/ of programs stat() way like crazy. |
19 |
> |
20 |
> |
21 |
> This has already been approached. See (http://lasr.cs.ucla.edu/conquest) for |
22 |
> details. Basically the mete-data and small files are stored in battery |
23 |
> backed RAM. They showed huge performance increases. |
24 |
|
25 |
-- |
26 |
gentoo-performance@g.o mailing list |