> > There are different kinds of "IOPS". The expensive ones are random > IOPS whereas sequential IOPS are much more efficient. The intention > of the SSD-based ZIL is to defer the physical write so that would-be > random IOPS can be converted to sequential scheduled IOPS like a > normal write. ZFS coalesces multiple individual writes into larger > sequential requests for the disk.
Yes I understand; but still isn't there a upperbond? If I would have the perfect synchronous ZIL load; and I would only have on large RAIDZ2 vdev in a single pool with 10TB, how would the system behave when it flushes the ZIL content to disk? > > Regardless, some random access to the underlying disks is still > required. If the pool becomes close to full (or has become fragmented > due to past activities) then there will be much more random access and > the SSD-based ZIL will not be as effective. Yes, I understand what you are saying but its more out of general interest what the relation is to the SSD devices vs. required (sequential) write bandwidth/IOPS. I can hardly imagine that there isn't one. Jeffry _______________________________________________ zfs-discuss mailing list zfs-discuss@opensolaris.org http://mail.opensolaris.org/mailman/listinfo/zfs-discuss
