Matthew Ahrens' Weblog

I thought that made for a good intro. One or two bits were slightly over the top I thought though. Still, a good mix of features, some implementation information and examples and contrasts (with older systems). Also made it clear that ZFS is far from incremental (evolutionary).

Two non trivial things I noticed:
1) Though simplicity is stressed, it would still be worth adding something like "with ZFS's natural simplicity, it's hard to setup a system poorly, both for reliability and performance".

2) No real distinction is made between low-end storage (say a 1U server with 1-2 hard discs) and a system with several TB of storage. Some workstation and HPC tasks can be very I/O intensive too. With some many "advanced" features being specified, it may come across as a little scary for more basic work. Putting it another way, I think you need to show that even very simple storage systems can have common issues (eg set root partition too small) and ZFS will wipe these out too.

Posted by Chris Rijk on September 14, 2004 at 04:01 PM PDT #

I would have liked to have seen the actual commands to type, eg zfscreate / zfsaddtopool , or whatever. Also some kind of idea on performance would be nice, if even only back of the envelope calcs. For instance: I have a U2 on my desk. It is connected via 100mbps ethernet to the file server. I run a database - where are the files written? Does ZFS "know" that my locally attached 15K cheetahs are the best place to write and read from?

Posted by PatrickG on September 14, 2004 at 06:08 PM PDT #

Chris, thanks for the advice. We'll think about emphasizing those benefits in future materials.

Patrick, we haven't finalized what the actual commands or performance will be, so detailing those probably wouldn't be very useful. Back of the envelope, ZFS performs very well -- on many tasks, it is much faster than traditional filesystems. Roughly speaking, you'll be able to do something like "zpool create disk1 disk2..." to create a storage pool, and "zfs create name_of_filesystem" to create a filesystem.

ZFS is a local filesystem. So if you want to write to disks on a different machine, NFS is still the right solution.

Posted by Matthew Ahrens on September 14, 2004 at 10:16 PM PDT #

I forgot to mention this previously - there was no mention of "DFS" in the article. Has this name been dropped in favour of ZFS? (that's great if so).

PS Good luck with the "expert exchange" thing. I wouldn't be too surprised if there's a number of questions like "how's it compare to reiser4" and so on.

Posted by Chris Rijk on September 15, 2004 at 07:06 AM PDT #

Could you please point me to the location of commentary of recently held chat session on ZFS(ie on September 15th 2004)? In your ZFS notes, it says that each user could have his own filesystem(scenario on the home directory server ). If there are several hundreds of users then there would be several hundred mount points. How are these mount points created/managed/mounted/umounted etc.?

Posted by Avis on September 20, 2004 at 05:05 PM PDT #

Avis, a transcript of the chat session is now available on the expert exchange website. We're still working on how best to help the administrator manage the potentially large number of filesystems available to them. One feature that ZFS will offer is the ability to define heirarchical groups of filesystems to manage any number of filesystems as one entity.

Posted by Matthew Ahrens on September 22, 2004 at 10:36 PM PDT #

Matthew, just a followup... today I was in Staples and they had 200GB IDE drives for $100 apiece. Figure a cheap $300 computer, Solaris 10, and 7x200GB drives (one hot spare, one for RAID5) - you have a full fledged terabyte ZFS box for about $1000.

Posted by PatrickG on October 14, 2004 at 09:39 PM PDT #

Matthew Ahrens' Weblog

Calendar

RSS Feeds

Search

Links

Navigation