On Wed, 26 Oct 2011 05:30:52 +1100
Justin Clift <jclift(a)redhat.com> wrote:
On the CloudFS Blog, in the initial "Why?" reading material, it says
this one of the things people look for is this:
* Performance and consistency for small writes in large files are
not reduced to near zero by doing read/modify/write on whole files.
That's not making sense to me. Specifically the "not" bit in it.
Why wouldn't people be wanting small write in large files to be fast?
If they're doing small writes in the middle of large files at all, then
they probably want those writes to be fast. With blob stores such as
S3 or Swift, performance for that I/O pattern is near zero because of
the need to do a read/modify/write. In GlusterFS/CloudFS/HekaFS,
"reduced to X by Y" does not apply.