Yes, but all of the NAS vendors have some sort of block storage (or SAN) on the back end. NetApps uses Fibre Channel arbitrated loop "SAN's" for the block access that their file system needs to access storage. So I don't ever think of this whole SAN versus NAS thing as relevant. You always have both file and block access needs. In the simplest case, NAS is the front end and SAN is the back end (and may be effectively hidden, as with NetApps), but there's no case where there's not some sort of block access going on behind a NAS.
-----Original Message----- From: owner-toasters@mathworks.com [mailto:owner-toasters@mathworks.com] On Behalf Of Glenn Walker Sent: Wednesday, February 01, 2006 7:57 AM To: Kevin H. Schoener; Michael Bergman; toasters@mathworks.com Subject: RE: Replacing our NetApp
To clarify this a bit further (it was a favored question of the interview teams at NetApp):
It's all about how the data is moved more than it is where it lives.
SAN is BLOCK LEVEL data transfer. NAS is FILE LEVEL transfer. In other words, SAN - iSCSI and FCP - both transfer data via blocks of information. NAS - NFS and CIFS - both transfer data via a file with a handle of some sort.
-----Original Message----- From: owner-toasters@mathworks.com [mailto:owner-toasters@mathworks.com] On Behalf Of Kevin H. Schoener Sent: Wednesday, February 01, 2006 3:01 AM To: 'Michael Bergman'; toasters@mathworks.com Subject: RE: Replacing our NetApp
Michael - here's my 2¢.
iSCSI and FCP are specifically SAN technologies and NFS is specifically not. The difference is not the fabric or packeting so much as it is where the file system lives.
NFS is NAS, so the file system is owned and controlled by the NAS device. In an iSCSI or FCP SAN, the file system is owned and controlled by the Server(s) to which the SAN is attached... Another way to think of it is that SAN technologies utilize device-type commands, whereas NAS technologies simply say "give me data or take the data" and it's up to the NAS device to decide where to put it...
I believe the point is that the NFS deployment on a NetApp filer takes FULL advantage of Data OnTap features and performance.
..okay, maybe I went longer than 2¢ worth... sorry
Kevin H. Schoener Lynx Technologies Inc.
KHSchoener@LynxTechnologies.NET http://www.LynxTechnologies.NET/
1576 Sweet Home Rd Suite 230C Amherst, NY 14228 716.636.5470 866.316.8599 fax 716.998.6065 cell
-----Original Message----- From: owner-toasters@mathworks.com [mailto:owner-toasters@mathworks.com] On Behalf Of Michael Bergman Sent: Wednesday, February 01, 2006 2:15 AM To: toasters@mathworks.com Subject: RE: Replacing our NetApp
What's wrong with SAN? iSCSI is most certainly SAN and has proven itself very well in the marketplace as well as being technically very
Fair point -- there's absolutely nothing wrong with SAN (= Storage Area Network, remember). I could argue, and keep that argument going, that if iSCSI is "SAN" (which I can agree that it can be) then so is NFS. Depending on how you build the infrastructure for the NFS server(s) and client(s).
But anyway, just to make things a bit clearer then on my part, stay away from non-encapsulated FC. iSCSI over Ethernet is fine :-) No one will care about FC anymore when iSCSI over 10 GbE is a commodity. It wont take *that* long before it is.
Cheers, /M
Is there any reason you kept talking about SANs? SAN has its place, especially if you need high end performance. But for a lot of applications NAS works just fine.
I agree. Use NFS. Configure it well. Build a "storage network" if you like, put in dedicated GbE switch(ws) (10 GbE?) with the Oracle server(s) and the NAS Filer(s) in it, make sure everything is Jumbo frames of course, tweak if for optimum performance as best possible, make sure traffic flows the way it should a.s.o. No client traffic on that net.
Use ONTAP 7.x (aggregates) and have enough spindles. Go for the smaller drives probably, not the 144G's, but the 72's
Most likely this will be Good Enough(TM) so you can stay away from complicated bug-ridden SAN setups that always are a nightmare to change config-wise. Especially if you need to change the config quite often in some way (= more than once a year...)
Just my 0.10 SEK worth
/M
On Wed, Feb 01, 2006 at 01:51:06PM -0800, Carter K. George wrote:
Yes, but all of the NAS vendors have some sort of block storage (or SAN) on the back end. NetApps uses Fibre Channel arbitrated loop "SAN's" for the block access that their file system needs to access storage. So I don't ever think of this whole SAN versus NAS thing as relevant. You always have both file and block access needs. In the simplest case, NAS is the front end and SAN is the back end (and may be effectively hidden, as with NetApps), but there's no case where there's not some sort of block access going on behind a NAS.
I think you are misguided.
To say there is no distiction between 2 competing technolgies because somewhere underneath they do something the same way is not sensible. I could dream up zillions of analogies, but one that may fit this case best is to say that there is no worhtwhile distiction between disk and tape. They both move magnetic media around under a head and read off a block of data. But thats not interesting when making a choice between whether to use a disk or a tape. The important thing is the attributes and cost of each technology as seen from the person (or system) using it.
And the salient distiction between SAN and NAS as stated before is that in one case the system sees blocks, talks via a block access protocol and runs its own filesystem, and in the other case the system sees files, and accesses the files via a file access protocol. The most important point is where the intelligence if the file system resides. With netapp, I get WAFL file system technology, for a SUN and SAN, I get UFS (maybe ZFS) technology.
I know which one I want.
Regards, pdg
--
See mail headers for contact information.
"Peter" == Peter D Gray pdg@uow.edu.au writes:
Peter> With netapp, I get WAFL file system technology, for a SUN and Peter> SAN, I get UFS (maybe ZFS) technology.
Don't forget VxFS, a very nice and manageable filesystem. Put it with VxVM and you've got alot of the OnTap 7 aggregate/flexvol stuff already there for you.
Peter> I know which one I want.
The one that makes the most sense for my needs. After my problems with OnTap 7.0.1R1 and quotas and qtrees earlier this week, I'm a bit leary of NetApp right now. OnTap 7G is still shaking out major bugs...
John
On Thu, Feb 02, 2006 at 08:59:12AM -0500, John Stoffel wrote:
"Peter" == Peter D Gray pdg@uow.edu.au writes:
Peter> With netapp, I get WAFL file system technology, for a SUN and Peter> SAN, I get UFS (maybe ZFS) technology.
Don't forget VxFS, a very nice and manageable filesystem. Put it with VxVM and you've got alot of the OnTap 7 aggregate/flexvol stuff already there for you.
Not standard in solaris, as far as I am aware. (correct me if I am wrong).
And you get some of the features of WAFL, but not others. And the ones you do get are implemented entirely differently, which has some impacts such as performance.
It appears from first glance that ZFS has nearly all the features of WAFL, at least conceptually if not in practice. Are they using netapp patents I wonder?
Peter> I know which one I want.
The one that makes the most sense for my needs. After my problems with OnTap 7.0.1R1 and quotas and qtrees earlier this week, I'm a bit leary of NetApp right now. OnTap 7G is still shaking out major bugs...
As others have said, you cannot criticize an OS when you are on a very early release that was superceded months ago. We have been on 7G now for months on all our filers without any issue at all. I will say that 7g when it was released was not as stable as 6, but thats hardly surprising given the magitude of the changes.
Regards, pdg
--
See mail headers for contact information.
It appears from first glance that ZFS has nearly all the features of WAFL, at least conceptually if not in practice. Are they using netapp patents I wonder?
ZFS is very similar in architecture to WAFL (In fact, I believe it was inspired by WAFL, and so is Reiser to a certain extent. At least Hans refers to it as a "well designed" filesystem.)
I like ZFS, but I wouldn't bet my paycheck with it vs. a Netapp Filer, version 7.0.1R or whatever:)
<propeller_hat> The primary percieved weakness of both is that neither is a "Cluster File System" (VxFS,Polyserve, OCFS etc.) , and those are very trendy these days, even though I don't see the point of doubling the locking overhead simply to have a high throughput file serving Cluster. (File Level *AND* block level locking seems to be a waste, pick one and use NFS or Fibre channel and your fancy CFS =)
So if you need multiple readers and writers accessing the same Filesystem via blocks, WAFL or ZFS won't work for you unless you use NFS, otherwise you're going to use the FCP/iscsi and a host based cluster FS, or some fancy Infinband Head clustering... but that has issues as well. </propeller_hat>
PS:
Regards, Max
Peter> I know which one I want.
The one that makes the most sense for my needs. After my problems with OnTap 7.0.1R1 and quotas and qtrees earlier this week, I'm a bit leary of NetApp right now. OnTap 7G is still shaking out major bugs...
As others have said, you cannot criticize an OS when you are on a very early release that was superceded months ago. We have been on 7G now for months on all our filers without any issue at all. I will say that 7g when it was released was not as stable as 6, but thats hardly surprising given the magitude of the changes.
Regards, pdg
--
See mail headers for contact information.
Don't forget VxFS, a very nice and manageable filesystem. Put it with VxVM and you've got alot of the OnTap 7 aggregate/flexvol stuff already there for you.
Not standard in solaris, as far as I am aware. (correct me if I am wrong).
Correct.
And you get some of the features of WAFL, but not others. And the ones you do get are implemented entirely differently, which has some impacts such as performance.
It appears from first glance that ZFS has nearly all the features of WAFL, at least conceptually if not in practice. Are they using netapp patents I wonder?
Certainly the concepts of how snapshots are managed and write aggregation/localization are similar. I think I've seen Sun engineers refer to the WAFL white paper in discussions on the technology.
The source for ZFS has been downloadable for a couple of months now at least. I'd assume Netapp has had a chance to look at it if they wanted to.
Peter> On Thu, Feb 02, 2006 at 08:59:12AM -0500, John Stoffel wrote:
> "Peter" == Peter D Gray pdg@uow.edu.au writes:
Peter> With netapp, I get WAFL file system technology, for a SUN and Peter> SAN, I get UFS (maybe ZFS) technology.
Don't forget VxFS, a very nice and manageable filesystem. Put it with VxVM and you've got alot of the OnTap 7 aggregate/flexvol stuff already there for you.
Peter> Not standard in solaris, as far as I am aware. Peter> (correct me if I am wrong).
Nope, it's not standard. It's an add on product.
Peter> And you get some of the features of WAFL, but not others. And Peter> the ones you do get are implemented entirely differently, which Peter> has some impacts such as performance.
I'm not sure what you're getting at here. As far as I can tell, VxFS on top of VxVM had *more* features in terms of sizing and such than did OnTap < 7G. It wasn't until we got FlexVols that you could actually shrink volumes, which VxVM and VxFS together could always do.
But I was admit they are targetting different audiences in some ways.
Peter> It appears from first glance that ZFS has nearly all the Peter> features of WAFL, at least conceptually if not in practice. Peter> Are they using netapp patents I wonder?
I dbout it...
Peter> As others have said, you cannot criticize an OS when you are on Peter> a very early release that was superceded months ago.
I can always criticize a release... even if I'm on an early release. The bug we hit was in regards to quotas and qtrees, something which OnTap has had for years. Basically, if you did:
qtree create qtree resize qtree delete qtree create
you could crash the filer and have it endlessly reboot, turn on quotas, crash, dump core, boot... not fun.
Peter> We have been on 7G now for months on all our filers without any Peter> issue at all. I will say that 7g when it was released was not Peter> as stable as 6, but thats hardly surprising given the magitude Peter> of the changes.
Sure, and I'm sure it will get better and better and better as time goes on. As I said elsewhere, we went with 7G because we were doing a full NetApp refresh last year and we wanted the FlexVols. Running into bugs was expected, just a bit disappointing at times.
But I still like and trust NetApps. Don't get me wrong.
John
-
Carter K. George -
Darren Dunham -
John Stoffel -
Max -
Peter D. Gray