On 6/25/12 6:04 PM, Earle, Erik wrote:
This is our current configuration (with 3.0.1). We are running up
against an upper limit with cache read-busy failures when we have many
simultaneous requests for the same object that I was hoping that using
a raw disk might fix (or give us a higher ceiling).
Are there other things to consider when getting read-busy failures?
Ah. Yeah, might help, not sure if it's going through a FS based image
anyways... (through the HV). For sure your best bet is always (and I
can't emphasize always enough) going to be direct IO to the raw device
(no HV, no FS).
-- Leif
