ZFS vs, NetApp's WAFL
UPDATE: i have corrected the glaring typo, its WAFL not AWFL and fine tuned other errors in the table. another new version availible now.
Someone on ZFS-discuss asked for a comparison of ZFS vs. NetApp's WAFL filesystem, so I spent a few minutes and created one. You can see it at
WAFL vs. ZFS pdf version
WAFL vs. ZFS html version
Someone on ZFS-discuss asked for a comparison of ZFS vs. NetApp's WAFL filesystem, so I spent a few minutes and created one. You can see it at
WAFL vs. ZFS pdf version
WAFL vs. ZFS html version
posted by jamesd_wi at Thursday, July 27, 2006
22 Comments:
Isn't it WAFL not AWFL?
This is highly suspect. Your not comparing WAFL vs ZFS, your comparing OnTap using WAFL vs Solaris using ZFS.
Some points that bother me...
OnTap doesn't support all the RAID levels you list, in fact it uses the one RAID level you don't list, RAID4. Also, you didn't point out that OnTap/WAFL supports Dual Parity, whereas ZFS doesn't (yet, its coming soon).
"Include Raw Devices"? I'm not sure what that even means. Your note in #2 makes no sense.
Filesystem for is "10's of gigabytes"? Try terabytes.
"Volume Support", "if the array supports iSCSI" needs more clarification than offered by #3. At the very least your saying iSCSI when you mean iSCSI and FCP.
OnTap is NOT limited to 255 snapshots and NetApp is the company that gave us clones. Its called SnapClone. Its one of the many features that ZFS copied.
How is the CLI for OnTap limited exactly? And where is this ZFS GUI?
This list is a nice attempt but full of gapping holes.
OnTap doesn't support all the RAID levels you list, in fact it uses the one RAID level you don't list, RAID4. Also, you didn't point out that OnTap/WAFL supports Dual Parity, whereas ZFS doesn't (yet, its coming soon).
okay the documentation I listed was vague on this I will change it when I next update it.
"Include Raw Devices"? I'm not sure what that even means. Your note in #2 makes no sense.
This means intergrating raw devices from other NAS, SAN or arrays into the filesystem. For example import 1TB of raw space via iscsi into the filesystem.
Filesystem for is "10's of gigabytes"? Try terabytes.
That was a direct quote from http://www.netapp.com/library/tr/3002.pdf
page 3, number 3 in the list, I looked for more specifications to do a closer comparison but google didn't get any.
"Volume Support", "if the array supports iSCSI" needs more clarification than offered by #3. At the very least your saying iSCSI when you mean iSCSI and FCP.
not sure what FCP is, I will fixed this with the next version.
OnTap is NOT limited to 255 snapshots and NetApp is the company that gave us clones. Its called SnapClone. Its one of the many features that ZFS copied.
okay, but this was also from the previous mentioned document on p3 section 2:
WAFL's primary distinguishing characteristic is Snapshots, which are read-only copies of the entire file system. WAFL creates and deletes Snapshots automatically at prescheduled times, and it keeps up to 255 Snapshots on-line at once to provide easy access to old versions of files.
if you have a link to a more accurate numbers please provide them.
How is the CLI for OnTap limited exactly? And where is this ZFS GUI?
the cli in WAFL is more spread out and difficult to learn and less flexible when you want to script it. ZFS has a much cleaner design and more understandable closer to the english language.
You can see the gui and how to use it at http://hell.jedicoder.net/?p=70
Note that the NetApp tech report is over 10 years old - might explain the capacity discrepencies and other differences.
Some corrections here with regards to WAFL:
1. RAID 1 is supported using SyncMirror feature. It'll mirror 2 RAID4 or RAID-DP together to form RAID 4+1 or RAID-DP+1
2. Silent data corruption is protected in WAFL with the "block checksum" feature. This feature has been available for years
3. WAFL supports Clone. The feature is called FlexClone.
All references to above can be searched on NetApp web site
2. Silent data corruption is protected in WAFL with the "block checksum" feature. This feature has been available for years
WAFL's checksum doesn't provide full protection, it only checksums the data that make it to the disk, it doesn't protect against a bad data getting sent to the drive. and it doesn't protect against errors happening after the data leaves the array. Bad cables, or drivers could effect data.
3. WAFL supports Clone. The feature is called FlexClone.
Yes I heard about it, but it is an extra cost feature.
I'm a fan of ZFS and have a little implementation experience, but I see that the comparison is one-sided and needs some balancing.
As far as I am aware, there is not yet any clustering solution for ZFS. If a node fails, you have to use your own methods and scripts to import your SAN-attached storage pool to another server and get back online. Any network connections/sessions to the file server would be lost and would have to be restarted. I'm sure people are working on at least active/passive failover for ZFS at Sun or Veritas, but nothing is available yet. ONTAP has active/active clustered failover for a WAFL volume. ONTAP GX has clustered filesystem support that improves their SPECnfs97 results to over a _MILLION_ operations per second. What's the best that a ZFS system on a Sun system can do? No one yet knows. A filesystem capable of "billions of gigabytes" is no good if a single front-end server can't handle any load comparable load.
I suspect that performance on comparable hardware would initally show a performance advantage to NetApp. Sun has yet to release any numbers for a ZFS-based product. I'd like to see SPECsfs97 numbers published for an X4600 8-way server with 64 GB RAM in front of lots of StorageTek 3510 storage to directly compare Sun X4600+ZFS+Solaris solution to a NetApp FAS6070A+WAFL+ONTAP solution. Until we see numbers published, I suspect Sun doesn't feel ready to compare its performance to others.
ONTAP has support for spare disks with automated recovery. With ZFS you have to detect a failure and initiate your own recovery scripts or manually respond to a failure. Spare disk support is almost ready for ZFS (planned in Solaris 10.3).
WAFL has been around for many many years with a proven base and support for commercial applications like Oracle and Exchange. ZFS is new and relatively experimental. There is still alot of development going on to add features to ZFS and fix bugs.
ONTAP knows how to effectively use NVRAM with WAFL to accelerate writes and ensure the last set of requests make it to storage during a power failure. It's unclear to me whether an acknowledged NFS write to a ZFS volume will make it to the disks if power abruptly fails.
The biggest advantages for ZFS: ZFS can cost you $0 in software costs to implement (and maintain!) and allows you to use any legacy hardware you want to build your systems. Multipathing support plus data checksums provide for better reliability and recoverability of data. WAFL is proprietary. OpenSolaris has a great Sun-supported development community, and one might even see a port of ZFS to other operating systems (DragonFly BSD is considering it).
Sun has surpassed Linux and Veritas in filesystem technology with ZFS and has caught up to WAFL with respect to snapshots (NetApp's biggest selling point), but Solaris+ZFS is far from being a clear winner compared to ONTAP+WAFL.
To summarize:
Cluster Failover - WAFL: Yes ZFS: No
Performance - WAFL: 1000000+ NFS ops / sec ZFS: Unknown
Hot Spares - WAFL: Yes ZFS: No(*)
Years in production - WAFL: 10+ ZFS: 0
Solaris/ZFS lacks one important snapshot feature that OnTap/WAFL provides. This feature is ability to QUICKLY and EASILY recover individual files from snapshots. This is extremely helpful when used for users home directories. I'm looking forward to the day when ZFS supports this feature.
With ZFS you have access to each snapshot in a form o file system. So restoring a file from any snapshot means just copy that file.
Lots of potential FUD here.
Lets face it, Immitation is the best form of flattery and it's very obvious that ZFS is a copy of WAFL.
ZFS is new, and has a couple of niche advantages, but the champ is still WAFL, hands down. Comparing SPECs on paper is hardly the way to go about testing these things, in the real world, WAFL is significantly more mature and stable than ZFS.
The two companies are partners as well as competitors, so keep this in mind:
I can run ZFS over NetApp LUNs which are embedded in WAFL. =)
WAFL's checksum doesn't provide full protection, it only checksums the data that make it to the disk, it doesn't protect against a bad data getting sent to the drive. and it doesn't protect against errors happening after the data leaves the array. Bad cables, or drivers could effect data.
This is an example of overselling something. ZFS and WAFL are both filesystems.. they're not HBA drivers, disk arrays or disk firmware. It's in these dark corners of the universe where "Silent" data corruption usually occurs, and the only way both file systems notice is when they read back the blocks in question.
Think about ZFS's END to END checksumming is basically a block level checksum as it relates to the filesystem (Which isn't "just on Disk"), just like NetApps.
A number of points in the comments are good. ZFS is a new file system and it will take some time to flesh out its feature set.
The other does conflate Data ONTAP (the file server architecture) with WAFL (the file system).
On data protection, WAFL prevents every form of data corruption that ZFS does and some that it doesn't. There is an excellent discussion of failures to watch out for in the uwisc IRON file systems paper. The mechanisms it uses are both hardware based (ECC protected data paths) and software based (end to end checksums with temporal signatures).
I enjoyed the comment about supporting "USB drives". Don't confuse an interface with a drive. True there are no USB disk drivers in WAFL and they aren't in ZFS either, rather if the operating system on which they were running supports them, then the file system supports them. Data ONTAP (the OS) does [i]not[/i] support using USB to talk to drives. (its too slow anyway).
Other people have mentioned FlexClone which is a writable snapshot.
WAFL's volume size limit is 16TB and on a single appliance you can have hundreds of volumes. The largest system, the FAS6070 supports over 500TB of disk attached to the system.
NetApp also sells their V-series which allows you to connect to any RAID array you'd like. Hitachi, IBM, HP, Etc. That is more expensive than the integrated RAID support but different strokes and all that.
I don't know that comparing them at this level is quite applicable but it has been entertaining to read.
--C
Your post said:
WAFL's primary distinguishing characteristic is Snapshots, which are read-only copies of the entire file system...
Actually, a SnapShot is a copy of the inodes only. In order to get a full copy of the actual data, you need to use SnapMirror to copy the data to another location. Ideally, this would be on a separate filer.
Please don't review filesystems you know nothing about and then point at marketing-grade literature as the source.
This comparison looks like someone is trying to make ZFS look better than its more mature parent, WAFL. It isn't. In fact, ONTAP supports RAID-1 and even RAID-10 can be configured, but it is recommended against because RAID-DP provides better protection than RAID-1, RAID-10, and RAID-5 for less cost and equivalent performance.
Furthermore what are ZFS's best practice on # of snapshots compared to performance impact? How about RAID-Z's performance impact? There is no info.
Whereas NetApp has many performance whitepapers. No. ZFS is a nice blueprint, but not ready for primetime.
NetApp has an appliance line that can virtualize raw device foreign storage into WAFL.
Each volume on NetApp is limited to 255 snapshots. However, there can be as many as 200 volumes per appliance.
I would be leery of anything opensource or recently announced from Sun. Sun has a way of making a splash announcement that fizzles in about a year. Whatever happened to Sun's N-series NAS arrays,
Java Desktop, or Javastation?
I would wait at least 3 years before I'd feel that ZFS has caught on.
In regards to CLI versus GUI. NetApp has a wonderful GUI for ONTAP. Ask Sun who won the desktop war based on their GUI, Sun or Microsoft. Evidently, most admins prefer an easy GUI over an easy CLI. Quite frankly, I don't see anything difficult about the NetApp CLI.
$ zpool status
That's English?
question?
what about snapshot replication. How
does ZFS handle this? You can do sync, semi-sync & async snapmirror on Netapp. Anything similar with ZFS?
You definitely forgot the row for "original work" vs. "transparent ripoff" and perhaps one for "astroturf free" vs. "fanbois astroturf everywhere" as well. Neither of those favor ZFS.
May be these comments are way off the mark in timelines.
However for the benefit of future readers.
NetApp filers have surely dominated the market in the past years that doesn't mean one can just ignore the revolutions that is happening at a cost that is affordable for any user who wants their data to be safe at a fraction of the cost of proprietary hardware provides.
ZFS might not yet have immediate numbers given that its so new. But it already
provides.
- RAID-Z2
- Hot Spares
and encryption support in progress,
And it already integrates all the features (barring few omissions) and more in progress.
And more its the Platform for General purpose Storage with Entireprise class protection and performance.
As far as calling products mature
will happen over time, and only because customers want to see some change in it.
Other then that I wouldn't call a sports car released this year less mature as it already integrates all the innovations that has happened over all the previous years.
Dear Platypus,
Would you call using technologies and platforms that existed before and innovating on top a 'transparent ripoff" ?
Then how about starting off making your own electricty of a different kind to power your machines ?
For that matter I really wonder if anyone should be using any materials or innovations in other fields at all.
I wonder why any one wants to build empires over free and opensource products like NFS or use free operating
systems to build software on top.
Why one should be using standard algorithms researched by someone else ?
What rights do they have to use data structures designed by others.
The point is innovation only happens over other past innovations and there is a huge difference between that and a 'transparent ripoff'
Boy, whoever wrote this
1. Needs to either update it with real information or take it down, as it's a regular google hit.
2. Needs to stop trying to bill himself as knowledgeable in this area.
3. Really should have tried harder.
Choose any one, or all three!
Some comments (and some comments on comments):
Actually, a SnapShot is a copy of the inodes only.
Actually, it's a copy of the root inode only :)
Max. volume size for WAFL is currently 16TB on all but 3 filer systems (250,270,2020), limited only by the max. size of the containing aggregate.
FCP means Fibre Channel (FCP is the name this protocol has within NetApp)
WAFL limits snapshots to 255 per volume, this is indeed a hard limit (I tried it)
SnapRestore is also an important WAFL feature, which restores a whole volume within seconds to the state of a snapshot. It can also copy back single files from a snapshot faster than a client computer can, because it copies the data directly on the filer instead of going through NFS/CIFS/FC/iSCSI. If you ever had to restore a 500gb LUN you will LOVE this feature :-)
Post a Comment
<< Home