AVS is a really bad idea

For those not familiar with AVS, you may (or may not) know it as the Sun StorageTek Availability Suite. The basic idea is that you can have two storage devices attached to different hosts that stay in sync. This product forms the core of the CDP (no, not that CDP) capabilities of the StorageTek and Sun Unified Storage Server product ranges.

Too bad it is completely and totally useless when using some of the standout features of Sun's current opus, ZFS.

Let's take one feature that shows ZFS in all its brilliance - resilvering. As detailed previously, doing this is going to be very bad for anyone who uses big disks and/or less than 10GbE connectivity between AVS nodes. Jim Dunham responded to this, saying he doesn't see a way of making AVS "smarter" or "ZFS aware". The man makes good points, but both he and the original post's author have totally missed the point.

Sun have done the wrong thing here with AVS:

• AVS is a paid-for add on for Solaris. Even though AVS is a Sun internal product, it's just as bad as VxFS or Veritas Volume Manager. Way to go, Sun!
• AVS in OpenSolaris is outdated. Really - why bother?
• AVS + ZFS just doesn't work, ever. It's a half-hearted effort that provides support for getting your data from one box to another, but no support from the vendor for failover unless you buy some black boxes. Take up religion if you hope to fail back gracefully.

ZFS is transactional and has intelligent operation management at its core. Send those transactional operations across the wire. A device resilvering should never have block rewrites replicated to the passive host and a scrub can happen in parallel to normal file system activities.

Take each of the above operations:

• A scrub completing with corrections successfully on the active host says nothing about disc consistency on the passive host. Begin the operation on both hosts.
• Even if you have two dedicated, nerdy and doomed-to-be-virgins-forever system administrators in your colo replacing disks in the active and passive nodes at the same time, there will be a difference between the two new disks as ZFS begins to resilver. Your disks are definitely inconsistant, unless you're using hardware RAID...WHICH MANAGES TO ROB YOU OF MANY OF THE BENFITS OF ZFS. For those in the audience that pledge an answer of "full AVS replication", see above. You fail.

If someone were to give me the means to satisfy my financial obligations and work on only ZFS, I would give you the solution to this problem inside of six months without a problem. Seriously Sun - if your (admittedly brilliant) software engineers can deliver the marvel that is RAID-Z in 599 lines, then I can do it.

For those who didn't read the intro of the blog, yes I am a software engineer and yes I am proficient in C, system programming and mathematics. My assertions are justified.

It is a horrible injustice that the greatest filesystem ever is treated in this way.