I-D ACTION:draft-ietf-ips-scsi-mib-04.txt

KRUEGER,MARJORIE (HP-Roseville,ex1) marjorie.krueger at hp.com
Wed Oct 30 11:08:03 PST 2002


* From the T10 Reflector (t10 at t10.org), posted by:
* "KRUEGER,MARJORIE (HP-Roseville,ex1)" <marjorie.krueger at hp.com>
*
I think we should do the same sort of thing in the SCSI MIB that we did in
the iSCSI MIB regarding counter items that may wrap (2 objects, c32 and c64,
the c32 is the lower 32 bits of the c64, c64 is optional).  

But why go back to bytes instead of Mbytes?  Maybe Kbytes - that's the
granularity of most file systems detailed file size.  Bytes seems too small
of a measure.

> -----Original Message-----
> From: Mark Bakke [mailto:mbakke at cisco.com] 
> Sent: Wednesday, October 30, 2002 11:00 AM
> To: KRUEGER,MARJORIE (HP-Roseville,ex1)
> Cc: 'Amir Shalit'; ips at ece.cmu.edu
> Subject: Re: I-D ACTION:draft-ietf-ips-scsi-mib-04.txt
> 
> 
> "KRUEGER,MARJORIE (HP-Roseville,ex1)" wrote:
> > 
> > > 1) Read/Write counters
> > >
> > > Counter32 (counting MB transferred) wrap at about 1000 hours on 
> > > 10Gbps links. Can we the MIB use a Counter64 instead?
> > 
> > We want to accommodate SNMPv1 agents (which can't implement 
> > Counter64), so we can add an optional counter64.  Thanks for the 
> > reminder!
> 
> Since we are back to an optional counter64, should it be in 
> bytes instead of MB?
> 
> > 
> > > 2) Virtualization
> > >
> > > An attempt was made to list all LU's which are part of a 
> LUN. In my 
> > > opinion, the ultimate mechanism to represent hierarchical 
> volumes is 
> > > via a volume manager MIB. For the time being it will be useful to 
> > > associate a {start LBA, end LBA} vector with each LU to allow for 
> > > most simplistic virtualization mapping.
> > 
> > I'm not sure what you mean.  The MIB lists all the LU's that are 
> > contained within a Target device, and then a table of all the LUN 
> > mappings for those LU's.  I don't really think of "LU's 
> being part of 
> > a LUN" - the MIB doesn't try to accommodate virtualization, 
> we agree 
> > that virtualization is better represented via another MIB.  
> This MIB 
> > is an attempt to represent the simple
> > (average?) SCSI device.  Section 3.4 states that this MIB 
> is not meant to
> > address virtual devices, merely the "visible SCSI 
> attributes" (what a host
> > will see).
> 
> I agree with Marj.  Note that even target mapping (e.g. map a 
> FC target to an iSCSI target) or LUN mapping are outside the 
> scope of a SCSI MIB.
> 
> > Regards,
> > Marjorie Krueger
> > Networked Storage Architecture
> > Networked Storage Solutions
> > Hewlett-Packard
> 
> -- 
> Mark A. Bakke
> Cisco Systems
> mbakke at cisco.com
> 763.398.1054
> 
*
* For T10 Reflector information, send a message with
* 'info t10' (no quotes) in the message body to majordomo at t10.org




More information about the T10 mailing list