Log in

No account? Create an account

!AX == good / ZFS on purple.the-7.net

This is the first year I skipped AX in God knows how many years, yet I feel all good and dandy and relaxed at home.  :D

Oh, by the way, purple.the-7.net has seen the light and converted to ZFS (except for /boot, which needs to be still in UFS because the boot loader is too stupid to handle ZFS yet).  I'm amazed how its performance rivals that of UFS2, even with all the extra features comes with it.  Gotta commend Sun Microsystems for releasing this amazing piece of code in open source.  :D

The LTO-2 drives in purple.the-7.net and seerajeane.astralblue.net (my bedroom computer) are awesome.  I'm copying/restoring purple:/pub onto seerajeane:/pub (which is on a dedicated RAID5 array and runs ZFS too), and the bottleneck is not the tape drive but the hard drive.  Modern tape drive technologies never cease to amaze me….  I mean, LTO-2 isn't even the newest generation of LTO technology (there's LTO-4 now, which boasts the throughput of 120MB/sec, which is faster than usual hard drives these days can handle)!

Speaking of ZFS, it's godly also because it spreads writes across the entire disk over time (it's called “wear leveling”) can detect bad sectors on the fly and adjust accordingly; it means it can run on top of a flash drive, whose memory cells (sectors) has a finite lifetime in terms of write operations.  My home router is now a Soekris net5501 box with a CompactFlash disk in it, and I can safely use read-write filesystems without having to worry about wearing the disk off in 10 months (it had happened to me once), thanks to ZFS.  ♥


I did some more digging and it seems I was wrong; wear-leveling is built not into ZFS but into flash drives themselves these days (in order to handle primitive filesystems such as FAT32, whose FAT and filesystem summary area need to be rewritten every time something is modified on the disk).  On the other hand, ZFS can detect write failures and push the sector out of service as necessary.
Ah, yeah, that makes more sense. Yeah, standard CF devices that present themselves like IDE/ATA devices will typically hide the wear-levelling logic on the inside. I know of flash filesystems, but they're more for direct NAND flash chip access.

I've so far liked ZFS. My only quibble with it is that it is presently impossible to use its ACL system to set up a completely wide open file share. (Which is bad, security-wise, but horribly convenient for home use.)