> From everything I've seen, an SSD wins simply because it's 20-100x the > size. HBAs almost never have more than 512MB of cache, and even fancy > SAN boxes generally have 1-2GB max. So, HBAs are subject to being > overwhelmed with heavy I/O. The SSD ZIL has a much better chance of > being able to weather a heavy I/O period without being filled. Thus, > SSDs are better at "average" performance - they provide a relatively > steady performance profile, whereas HBA cache is very spiky.
This is a really good point. So you think I may actually get better performance by disabling the WriteBack on all the spindle disks, and enabling it on the SSD instead. This is precisely the opposite of what I was thinking. I'm planning to publish some more results soon, but haven't gathered it all yet. But see these: Just naked disks, no acceleration. http://nedharvey.com/iozone/iozone-DellPE2970-32G-3-mirrors-striped-WriteThr ough.txt Same configuration as above, but WriteBack enabled. http://nedharvey.com/iozone/iozone-DellPE2970-32G-3-mirrors-striped-WriteBac k.txt Same configuration as the naked disks, but a ramdrive was created for ZIL http://nedharvey.com/iozone/iozone-DellPE2970-32G-3-mirrors-striped-ramZIL.t xt Using the ramdrive for ZIL, and also WriteBack enabled on PERC http://nedharvey.com/iozone/iozone-DellPE2970-32G-3-mirrors-striped-WriteBac k_and_ramZIL.txt This result shows the WriteBack enabled makes a huge performance difference (3-4x higher) for writes, compared to the naked disks. I don't think it's because an entire write operation fits into the HBA DRAM, or the HBA is remaining un-saturated. The PERC has 256M, but the test includes 8 threads all simultaneously writing separate 4G files in various sized chunks and patterns. I think when the PERC ram is full of stuff queued for write to disk, it's simply able to order and organize and optimize the write operations to leverage the disks as much as possible. _______________________________________________ zfs-discuss mailing list zfs-discuss@opensolaris.org http://mail.opensolaris.org/mailman/listinfo/zfs-discuss
