On Wed, 13 Dec 2000, Robert Collins wrote:
> regarding fragmenting the object store - yes I agree. Sending out
> a hit of a 1 MB file from 1 different locations on the squid disks
> isn't ideal.
[ assuming you meant "100 different locations" ... ]
I would be cautious about this. If many requests for different objects
go to the same disk, then it does not matter much whether subsequent
requests for parts of the same object go to the same place on disk.
The presence of other requests and elevator (or SCAN) seek
optimization will most likely eliminate all advantages of storing
related chunks close to each other. Fortunately, with enough requests
in the queue, the seeks will be minimal thanks to the seek
optimization policy.
Alex.
Received on Wed Dec 13 2000 - 11:31:41 MST
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:13:04 MST