64-bit SCSI LUN addressing on Linux

64-bit SCSI LUN addressing on Linux

[Martin W. Schlining III]

Does Linux support 64-bit SCSI LUN addressing? Suppose I have a system 
that can create SCSI LUNs that are greater than 2TB using a block size 
of 512 bytes and can accept 16 byte CDBs, can I use this LUN with a 2.6 
series Linux kernel?

If so, are there any filesystems (ext3, XFS, etc. ) that can use a LUN > 
2TB?

Would I need any additional tools?

Would I need a 64-bit platform?

Re: 64-bit SCSI LUN addressing on Linux

[Christoph Hellwig]

On Thu, Jun 30, 2005 at 12:01:09PM -0400, Martin W. Schlining III wrote:
> Does Linux support 64-bit SCSI LUN addressing? Suppose I have a system 
> that can create SCSI LUNs that are greater than 2TB using a block size 
> of 512 bytes and can accept 16 byte CDBs, can I use this LUN with a 2.6 
> series Linux kernel?

Yes.

> If so, are there any filesystems (ext3, XFS, etc. ) that can use a LUN > 
> 2TB?

XFS has been tested heavily with big luns, but the others should work
aswell.

> Would I need any additional tools?

no.

> Would I need a 64-bit platform?

For filesystem sizes < 16TB you can use 32bit platforms, above that you
need a 64bit kernel.

Re: 64-bit SCSI LUN addressing on Linux

[Douglas Gilbert]

Martin W. Schlining III wrote:
> Does Linux support 64-bit SCSI LUN addressing?

No. Linux supports 32 bit LUNs.

> Suppose I have a system 
> that can create SCSI LUNs that are greater than 2TB using a block size 
> of 512 bytes and can accept 16 byte CDBs, can I use this LUN with a 2.6 
> series Linux kernel?

There is no relationship between 64 bit LUNS (which,
for example, support a hierarchy of up to 4, 16 bit LUNs)
and the ability of a logical unit (LU) (e.g. a disk) to
support 64 bit logical block addressing (LBA).

Christoph's reply more accurately addresses the question
I think you were trying to ask...

In your title perhaps you could: s/LUN addressing/LBAs/
LUNs are about device addressing; LBAs are about block
(or sector) addressing on a block device.


Doug Gilbert

Re: 64-bit SCSI LUN addressing on Linux

[Christoph Hellwig]

On Thu, Jun 30, 2005 at 12:57:00PM -0400, Douglas Gilbert wrote:
> Martin W. Schlining III wrote:
> >Does Linux support 64-bit SCSI LUN addressing?
> 
> No. Linux supports 32 bit LUNs.
> 
> >Suppose I have a system 
> >that can create SCSI LUNs that are greater than 2TB using a block size 
> >of 512 bytes and can accept 16 byte CDBs, can I use this LUN with a 2.6 
> >series Linux kernel?
> 
> There is no relationship between 64 bit LUNS (which,
> for example, support a hierarchy of up to 4, 16 bit LUNs)
> and the ability of a logical unit (LU) (e.g. a disk) to
> support 64 bit logical block addressing (LBA).
> 
> Christoph's reply more accurately addresses the question
> I think you were trying to ask...

Yeah, people tend to say LUNs when they mean storage devices a lot,
so I've started to look over the slight misuses if things make sense
otherwise.

Re: 64-bit SCSI LUN addressing on Linux

[Bryan Henderson]

>Yeah, people tend to say LUNs when they mean storage devices a lot,
>so I've started to look over the slight misuses if things make sense
>otherwise.

We all have to do that, because the vast majority of people who are not 
storage engineers, and a disturbing portion of people who are, mean LU (or 
storage device) when they say LUN.  But we shouldn't compound the problem 
by not pointing out the mistakes; people deserve to know there's a clearer 
way to say what they're trying to say.  In many contexts, I have been 
genuinely confused as to whether a person was talking about a LUN or  a 
LU.  "LUN discovery," for example is a much simpler problem than the more 
general problem of LU (or volume or device) discovery.  I end up having to 
do a meta-analysis of which language the author is most likely to be 
speaking.  In the current discussion, we have the phrase "64-bit SCSI LUN 
addressing," which does sound a whole lot like it's about 64 bit LUNs. 
Only when you look closely do you see that decoded the other way, it makes 
more sense.

--
Bryan Henderson                     IBM Almaden Research Center
San Jose CA                         Filesystems