On Fri, Oct 10, 2008 at 11:27:07AM -0700, Marion Hakanson wrote: > Yes, this advice was unexpected to me. Slicing up a spinning disk drive > is going to cause lots of extra seeks, but I'm having a hard time imagining > that an SSD is going to even notice that it's been sliced up. > > So, enlighten me: How can slicing an SSD into logical partitions slow > things down, anyway?
Hey Marion, That's a reasonable question. Here are some issues that my experimentation with SSDs has hilighted: The controller on the SSD and the individual NAND flash chips can handle only a certain number of IOPS. Depending on the device reads and writes may or not be prioritized in the way that you'd like. For example, if you have an SSD that's acting both as ZIL and L2ARC, it's likely that the large writes to populate the L2ARC would impact the latency of (higher priority) writes to the ZIL thus adversely impacting performance. Metadata used by SSDs internally can be impacted by multiple consumers. In ZFS we attempt to maintain locality, but if a device is being used in two capacities ZFS no longer has that global perspective of how the device is being used. I hope that makes thing clearer. It may be that for a particular use case, the use of different slices don't interact, but in general I'd argue that that design is suboptimal for performance. Adam -- Adam Leventhal, Fishworks http://blogs.sun.com/ahl _______________________________________________ storage-discuss mailing list [email protected] http://mail.opensolaris.org/mailman/listinfo/storage-discuss
