Solaris 10: tips and tricks for system administration - ZFS

Местные новости: ZFS Performance Talk

nospam@example.com (Gleb Reys) — Tue, 03 Apr 2007 12:42:03 +0400

Tim Foster had sent an email to IOSUG mailing list yesterday, that tomorrow, at 7pm, Roch Bourbonnais will present a tech talk on ZFS Performance.

Here's the announcement:

So, amazingly short notice here, but we're going to hold a talk on ZFS Performance this Wednesday.

Roch Bourbonnais, an expert on kernel performance is visiting Dublin on Wednesday and we thought we'd take the opportunity to hear more about ZFS internals & performance from him.

Roch has been examining ZFS in great detail during it's development with a view to it's characteristics on real-world workloads. If you're interested in ZFS, or have heard about it but want to know more, you really should come along on Wednesday.

More information on Roch's background and experiences with ZFS so far at http://blogs.sun.com/roch/

Again, our apologies for the short notice.

All the details are on the official IOSUG8 invitation page.

ZFS administration guide: now in Russian

nospam@example.com (Gleb Reys) — Wed, 07 Mar 2007 01:43:04 +0300

At last, I see some real progress with available translations of excellent Solaris docs!

For instance, ZFS administration guide in Russian was posted on the OpenSolaris/G11 (Globalization) Consolidation page just a few days ago.

It really is great to see great translations like this completed. Well done, guys!

ZFS: commands emulation

nospam@example.com (Gleb Reys) — Wed, 15 Mar 2006 01:20:27 +0300

I'm considering getting myself an x86 server with Solaris for home use, and today I've decided to play around with ZFS on my old Sparc-box.

I believe those of you who are just getting started with ZFS will really like the -n command line option for zfs and zpool, which allows you to see what result would a certain command yield, while there is nothing actually done to your disks.

So, if you're thinking of creating a new storage pool, you don't have to actually create it just to see how it will look - you can use -n option instead:

bash-3.00$ zpool create -n mypool raidz /export/vdev/d1 /export/vdev/d2 /export/vdev/d3 /export/vdev/d4

would create 'storage' with the following layout:



             mypool

                raidz

                   /export/vdev/d1

                   /export/vdev/d2

                   /export/vdev/d3

                   /export/vdev/d4

The command will still be validated - its syntax will be checked, plus you'll have the existance of the disks verified as well. So, if you don't have one of the devices in your system, you might see something like this:

bash-3.00$ zpool create -n mypool raidz /export/vdev/d1 /export/vdev/d2 /export/vdev/d3 /export/vdev/d4

cannot open '/export/vdev/d1': No such file or directory

ZFS: filesystem properties

nospam@example.com (Gleb Reys) — Tue, 24 Jan 2006 00:28:45 +0300

There is quite a number of ZFS filesystems properties, and I will cover most useful ones today.

I'll begin with refreshing your knowledge a bit. As I've told before, I'm going to create a number of ZFS filesystems on my server for the next few blog entires on ZFS. So, right now I have the following:

CODE:

bash-3.00# zfs list
NAME                   USED  AVAIL  REFER  MOUNTPOINT
stock                 2.40G  31.1G  8.50K  /stock
stock/oracle          2.32G   693M  2.32G  /stock/oracle
stock/try                8K   512M     8K  /try

To get a value of some ZFS property, we normally use a command like this:

CODE:

bash-3.00# zfs get used stock/try
NAME PROPERTY VALUE SOURCE
stock/try used 8K -

But such output isn't always convenient for us - it's formatted this way to make the data more human readable. But if we're going to use zfs get to obtain some ZFS parameters from our scripts, we really don't need anything fancy. So we're probably better off using a special -H command line option:

CODE:

bash-3.00# zfs get -H used stock/try
stock/try used 8K -

One more thing - a useful, yet very obvious feature which allows you specifying few properties names in one command line:

CODE:

bash-3.00# zfs get -H used,available,mounted stock/try
stock/try       used    8K      -
stock/try       available       512M    -
stock/try       mounted yes     -

Okay, now let's move on and look at some useful ZFS filesystems properties.

For instance, the mountpoint property name speaks for itself. By default, the filesystem which you change the mountpoint for, will be automatically remounted at a new location:

CODE:

bash-3.00# zfs set mountpoint=/younameit stock/try
bash-3.00# zfs list
NAME                   USED  AVAIL  REFER  MOUNTPOINT
stock                 2.40G  31.1G  8.50K  /stock
stock/oracle          2.32G   693M  2.32G  /stock/oracle
stock/try                8K   512M     8K  /younameit
bash-3.00# df -k | grep stock
stock                35112960       8 32597477     1%    /stock
stock/oracle         3145728 2436188  709540    78%    /stock/oracle
stock/try             524288       8  524280     1%    /younameit

One more potentially useful option is an on-disk compression. It happens on the fly, and it is planned to eventually have a number of compression methods available to a ZFS administrator, but at the moment you can use only one method - lzjb.

Still, results are impressive:

CODE:

bash-3.00# ls -l /younameit/access_log
total 3082
drwxr-xr-x   2 root     sys            3 Jan 23 21:59 .
drwxr-xr-x  44 root     root        1024 Jan 23 21:45 ..
dr-xr-xr-x   3 root     root           3 Jan 23 22:04 .zfs
-rw-r--r--   1 root     root     8904090 Jan 23 21:59 access_log
bash-3.00# zfs list
NAME                   USED  AVAIL  REFER  MOUNTPOINT
stock                 2.41G  31.1G  8.51M  /stock
stock/oracle          2.32G   693M  2.32G  /stock/oracle
stock/try             1.51M   510M  1.51M  /younameit

No wonder the compression ratio is so good for this filesystem:

CODE:

bash-3.00# zfs get compressratio stock/try
NAME PROPERTY VALUE SOURCE
stock/try compressratio 5.63x -

Unfortunately, only few types of files can be so wonderfully compressed, and text files (for instance, my Apache2 logs) belong to them too. Executable files, of course, will compress with a much lower ratio.

I think that should do for today's entry. Let me tell you right now: I have a whole series of ZFS blog entries planned ahead, and that's the reason I haven't covered some of the simpler ZFS filesystem options yet. I promise to write more in the nearest few days.

ZFS: managing filesystems

nospam@example.com (Gleb Reys) — Wed, 18 Jan 2006 00:21:19 +0300

Hi!

Unbelievable - this turns out to be the very first entry of mine in 2006! I couldn't postpone new entires anymore, so here's one for you :)

It's been quite a while since I've cleared one of internal disks in my Netra t105 to bring it under ZFS control. As a result, I now have a 33Gb zfs-pool to experiment with. Today I had some spare time, so I've decided to share the very basics of managing ZFS filesystems.

So this is the pool I have:

CODE:

bash-3.00# zpool list
NAME SIZE USED AVAIL CAP HEALTH ALTROOT
stock 33.8G 2.40G 31.4G 7% ONLINE -

And like I said, it has only one drive in it:

CODE:

bash-3.00# zpool status
  pool: stock
state: ONLINE
scrub: scrub stopped with 0 errors on Tue Jan 17 21:00:45 2006
config:

        NAME        STATE     READ WRITE CKSUM
        stock       ONLINE       0     0     0
          c0t1d0    ONLINE       0     0     0

By default, when you create a zfs-pool, all of its disk space is represented as a single filesystem with the same name. This filesystem also gets a mountpoint with the same name and is automatically mounted off /. So, after doing a zpool create (here is my example) you can immediately start working with your newly made filesystem.

At the moment I've got only 1 additional filesystem created, I use it for Oracle 10g. So here's how it looks:

CODE:

bash-3.00# zfs list
NAME                   USED  AVAIL  REFER  MOUNTPOINT
stock                 2.40G  31.1G  8.50K  /stock
stock/oracle          2.32G   693M  2.32G  /stock/oracle

According to ZFS concepts, all the newly made filesystems may use all the available disk space of the zfs-pool they belong to. So, when I create another filesystem in my stock pool, this filesystem has more than 30Gb available to it:

CODE:

bash-3.00# zfs create stock/try
bash-3.00# zfs list
NAME                   USED  AVAIL  REFER  MOUNTPOINT
stock                 2.40G  31.1G   9.5K  /stock
stock/oracle          2.32G   693M  2.32G  /stock/oracle
stock/try                8K  31.1G     8K  /stock/try

To make sure that some of your filesystems don't eat up all the available disk space, you have to limit them. And it's very easily done so:

CODE:

bash-3.00# zfs set quota=512m stock/try
bash-3.00# zfs list
NAME                   USED  AVAIL  REFER  MOUNTPOINT
stock                 2.40G  31.1G   9.5K  /stock
stock/oracle          2.32G   693M  2.32G  /stock/oracle
stock/try                8K   512M     8K  /stock/try

As you have probably guessed, quota is the parameter name, while set is a keyword to alter parameters values. There is quite a number of ZFS filesystem parameters, and most of them you can modify, but there are also some which are read-only. Here's an example of such a parameter, and you can see what happens if you try modifying them:

CODE:

bash-3.00# zfs get mounted stock/try
NAME PROPERTY VALUE SOURCE
stock/try mounted yes -
bash-3.00# zfs set mounted=on stock/try
cannot set mounted property: read-only property

You can find the full list of ZFS filesystem parameters on the man page for zfs, as for me - I'll show you only 1 more parameter today, and a very useful one in my opinion: a mount point. As all the rest things about ZFS, this parameter is very easy to change, and you even have the filesystem automatically remounted for you:

CODE:

bash-3.00# df -k | grep ^stock
stock                35112960       8 32597478     1%    /stock
stock/oracle         3145728 2436188  709540    78%    /stock/oracle
stock/try             524288       8  524280     1%    /stock/try
bash-3.00# zfs set mountpoint=/try stock/try
bash-3.00# df -k | grep ^stock
stock                35112960       8 32597478     1%    /stock
stock/oracle         3145728 2436188  709540    78%    /stock/oracle
stock/try             524288       8  524280     1%    /try

I left the df output intentionally here, to show that there were no additional manipulations made to the try filesystem - it really is this easy!

I guess that's enough for today - so next time I'll tell you about few more useful ZFS filesystem parameters. Good luck with your experiments!

ZFS features: RAID-Z

nospam@example.com (Gleb Reys) — Tue, 29 Nov 2005 03:16:00 +0300

I started slowly building up my knowledge of ZFS, and have decided to start with the most tasty things. One of them is a RAID-Z approach.

Jeff Bonwick - a ZFS author and one of newly created OpenSolaris ZFS community leaders, wrote a wonderful article on RAID-Z where he thoroughly explains what's new and cool about RAID-Z compared to RAID-5. Please read this masterpiece here.

In short, here's what's cool about RAID-Z:

1) A traditional weakness of parity-based RAID schemes - write hole - is fixed

2) There's quite a performance improvement because now we have dynamic length stripes - any small amount of data will be treated as a data block of a full-stripe size, and therefore it will be immediately and fully written onto disks.

3) Data integrity improved - in addition to traditional XORs to validate data based on parity knowledge, each RAID-Z block is validated with a 256-bit checksum.

4) ZFS has a self-healing functionality built-in. If one of data disks returns corrupted data, ZFS restores the correct data with parity, compares checksums to validate the result, returns this correct data to the disk operation which requested it, and fixes corrupted data on the disk which gave us the corrupted data. Must be really useful!

That's all for now. Soon there will be new ZFS presentations - and I hope they will have even more information on RAID-Z.

When ZFS will be available for Solaris 10

nospam@example.com (Gleb Reys) — Sat, 26 Nov 2005 03:06:19 +0300

I've been asked yesterday if it's known yet about the date when ZFS is expected to make its way into Solaris 10. There is no exact answer, but the official ZFS FAQ page hosten at OpenSolaris.org is and will probably be the most reliable source of such information.

This is the answer currently found in ZFS FAQ:

ZFS should be available in Solaris 10 Update 2, which is scheduled to ship in Q2CY06. All such plans are tentative, however, and subject to change.

ZFS: the first encounter

nospam@example.com (Gleb Reys) — Thu, 24 Nov 2005 18:03:05 +0300

I just couldn't wait any longer, and decided to try something with ZFS. And what do you know? I've found a very useful advice from Ben Rockwood, which allows us use regular files created with mkfile as virtual disks for ZFS. VERY useful, especially when you really want to play with such a great technology and maybe try various configurations, but there are no spare physical disks for such experiments.

I will, of course, read and understand all the inner workings of ZFS later, but the first impression of it is this - too easy to be true! :)

Here's what I did:

CODE:

bash-3.00# mkdir /export/vdev
bash-3.00# mkfile 128m /export/vdev/d1
bash-3.00# mkfile 128m /export/vdev/d2
bash-3.00# zpool create try /export/vdev/d1 /export/vdev/d2
bash-3.00# zpool status
  pool: try
state: ONLINE
scrub: none requested
config:

        NAME               STATE     READ WRITE CKSUM
        try                ONLINE       0     0     0
          /export/vdev/d1  ONLINE       0     0     0
          /export/vdev/d2  ONLINE       0     0     0
bash-3.00# zpool list
NAME                    SIZE    USED   AVAIL    CAP  HEALTH     ALTROOT
try                     246M   32,5K    246M     0%  ONLINE     -

After this the newly made filesystem automatically became available (it's a partial output of df -k):

CODE:

Filesystem kbytes used avail capacity Mounted on
try 235520 3722 231771 2% /try

Enough for the moment! Now the next challenge will be freeing up one of the two physical disks I've got in my Netra t1, so that I could bring it under ZFS control.

ZFS: feel the difference

nospam@example.com (Gleb Reys) — Mon, 21 Nov 2005 14:35:09 +0300

While reading Sun blogs today, I've spotted a nice article: Tim Foster has demonstrated how many things get vastly simplified with ZFS, in his blog entry ZFS, is that it?

I can't wait to try something ZFS myself! Unfortunatelly, got no spare time this week, and it's not clear where I might have some.

ZFS

nospam@example.com (Gleb Reys) — Fri, 18 Nov 2005 11:50:08 +0300

At last, it's here! ZFS is now publicly available in the latest build of OpenSolaris :)

It really is a pity that my home server has only 2 HDDs, which, of course, won't allow me to see and show all the features of ZFS... But some things, I'm sure, can be tried even on such a small scale ;) So I'll be sure to post more on that later on.

As for now - I advise you all have a look at ZFS basics presentation by Dan Price