CURRENT MEETING REPORT
Reported by David Wihl, Digital Equipment Corp.
Minutes of the Redundant Arrays of Independent Disks BOF (RAIDMIB)
The RAID MIB BOF met at the Dallas IETF on 4 December 1995. The
meeting was chaired by Mark Johnson, Symbios Logic, Inc.
(mark.johnson@symbios.com)
Agenda:
Introduction
Technology - RAID and RAID mgmt
History
Vendor Proprietary Activities
Discussion of Goals
Discussion of Charter Proposal
Mark provided a summary of RAID technology and RAID management
technology, and its relationship to enterprise-level system
management. A discussion of the general attributes of RAID
management, and its complexity relative to disk management,
followed. A major part of the problem statement is the
development of "Network Attached" storage subsystems (whatever
that ends up actually becoming) which must by definition be
"network manageable."
At present, network-enabled RAID management is a rather chaotic
collection of vendor unique solutions. The following group of
RAID vendors have developed proprietary MIB definitions to
support the management of their own devices:
- Symbios (NetWare only)
- Adaptec
- BusLogic (decided not to release?)
- Storage Dimensions
- Compaq
- Conner
- DG (Clariion)
- EMC
- Pacific Micro Data
- AMI?
By late 1994, a demand seemed to be emerging for the development
of a standard for, at minimum, network monitoring of RAID and
other storage-array technologies. Discussions on several Usenet
newsgroups and mailing lists led to the following informal
meetings being held in 1995 to discuss the possibility for
agreement on a common management framework.
Spring +95 N+I: Attended by technical representatives from
ArrayTech, Buslogic, Conley, EMC, Hitachi, Novell, Peer, SNMP
Research, Symbios, Skylight, Trimm. Discussions included
presentation of two different early proposals from vendors, based
on their proprietary work in existence or under development.
Mid-year 95: The RAIDMIB discussion list was created, and remains
available (subscription to raidmib-l-request@netx.com. Several
proposals for consensus lists of management variables were
discussed in the months that followed.
Fall +95 N+I: BOF, Atlanta N+I. A rough draft of the charter
proposal presented at the Dallas IETF was developed by consensus
of those present. Attendance: Adaptec, Bell South, Conley, DEC,
Fujitsu, Nabisco, RAID Advisory Board (RAB), Symbios, S/W
Technologies Gr., TI.
Major goals for the proposed RAIDMIB WG:
- standardize basic management variables
- standardize configuration description
- reduce "reinvention of the wheel"
- cooperate with RAID Advisory Board(RAB), an industry group of
RAID providers and users, and other standards groups to develop a
common language for management of array storage.
Additional goals of WG:
- produce a single standard MIB and MIF, with both management
frameworks deriving their structure from a common, if perhaps
abstract model.
Overriding principle : " Enough information to fix it when it is
broke."
Relationships with other work in progress: Is there a dependency
on the Entity MIB (containment relationship and logical
dependencies)?
a: Unknown at this time. If it+s present we+ll use it. Visit
Entity MIB WG, Tuesday 1:00 PM. Promoters of the RAIDMIB activity
need to be aware of the Entity MIB WG+s activity.
We discussed the success criteria for the activity (i.e., how
will a RAIDMIB WG know when to stop):
- notify of fault with enough info to find the part
- current status
- minimal physical configuration (show inter-relationships
between objects)
- e.g. perhaps show S/N of drive to know that drive has
moved
- Provide minimal performance statistics, with as precise
definitions as possible (e.g. bytes read/written, ,etc.) - where
you measure it, is useful. It seems important to provide well
defined information at well defined measurement points in order
to make it useful across multiple vendors and products.
- Provide logical mapping to physical configuration (this is
where the relationship to the Entity MIB work arises).
- Define something that is small, useful, and completed more
quickly, rather than complete in every detail. Mechanisms already
exist for vendor augmentation of a standard MIB.
Technical Issues Discussion:
Provisioning was raised as issue: Instrumentation should advise
of potential for expansion, and notify upon new addtions. It was
suggested that we should best avoid READ/CREATE row objects as
opposed to READ/WRITE.
q: Do we want multiple RAID subsystems per host? This could have
implications for MIB naming.
a: Yes, this is current state of the art.
q: Both in MIB and MIF: how do we handle instrumenting two
different vendors RAID subsystems on the same host?
a: Unknown. To be dealt with.
q: Do we want to use SCC? (The SCSI Controller Command set, the
ANSI-standard model definition for storage arrays, developed
principally by George Penokie of IBM)
a: No, it is too complex. This conflicts with the objective of
producing something small and useful, quickly.
q: Do we want to use defined RAID (0, 1, 0+1, 3, 5) levels?
a: Even in the RAB, RAID levels are going away (edition 6 of RAID
Book). We will study this. Possibilities include: Redundant, etc.
For many purposes, the RAID levels, which describe the data
mapping algorithm, do not convey essential information. Knowing
that redundancy exists, and which physical drives provide
redundant coverage, may be more important.
q: Should automated data rebuild rates be managed as a part of
this discussion? If so, how?
a: This was agreed to be a useful criterion. We will need to
agree on a collection of "levels" of rebuild rate, based upon
estimated impact to normal operations. Jeff Case suggested using
a requested (r/w) value vs granted (r/o) value, to allow vendors
to define their specific implementation of rebuild rates.
The charter was re-visited. Cooperation with the DMTF was re-
instated after deletion subsequent to the September N+I meeting.
The direct references to SCSI devices were removed, in
recognition of other data bus technologies.
Before the meeting concluded, the following questions were asked
and answered:
1. Is there sufficient interest among the providers of RAID
technology in incorporating a standard MIB in their products
to justify the investment in its development?
2. Is there sufficient interest among the consumers of RAID
technology in deploying products incorporating a standard
MIB to justify the investment in its development?
3. Is there sufficient interest among the attendees in
participating in the definition of a standard MIB (i.e., are
there sufficient willing workers?).
The answer to all three questions was "yes."
Redundant Arrays of Independent Disks
"RAIDMIB" Charter Proposal
Chair(s)
TBD
Network Management Area Director
Deirdre Kostick <kostick@qsun.ho.att.com>
Advisor
TBD
Mailing List Information
raidmib-l@netx.com (to subscribe, send "subscribe raidmib-l" in
the body of a message to raidmib-l-request@netx.com)
Description of Working Group
The RAID MIB Working Group will produce a document that defines MIB
objects for use in monitoring and (possibly) controlling RAID (Redundant
Arrays of Independent Disks) systems and subsystems.
Three sets of managed objects are anticipated:
The first set of managed objects allow the remote monitoring of RAID
systems.
The second set of managed objects allow the remote control of RAID
systems.
The third set of managed objects allow the remote configuration of RAID
systems.
The document resulting from the efforts of the Working Group is intended
to be submitted for consideration by the IESG at the entry to the IETF
standards track, i.e., as a Proposed Standard.
History, Goals and Milestones:
March 29 1995:
First meeting at Trimm Technologies to determine interest in the
concept.
September 27 1995:
First Interop BOF meeting in Atlanta.
December, 1995:
Organizational BOF at IETF, Dallas, Texas
January, 1996:
Working Group Chartered (?)
February, 1996:
Post initial draft MIB to Internet Drafts
March, 1996:
Working Group meetings, Los Angeles, CA
May - August 1996:
Additional meetings as necessary and appropriate
September, 1996:
Submit the RAID MIB to the IESG for consideration as a Proposed
Standard.
Current Internet-Drafts:
None at this time.
Requests for Comments:
None at this time.
It is the intent of this proposed Working Group to produce a standard
model defining management of array devices. It is intended that this
work will be done in collaboration with other bodies. Specifically,
this work to define an appropriate MIB within the IETF will parallel
efforts within the Desktop Management Task Force (DMTF) to define a MIF.
It is intended that the MIB and MIF definitions will be closely aligned.
In addition, active participation in the Working Group by designated
representatives of the RAID Advisory Board will be sought and
encouraged, in order to help ensure that the descriptive model that is
developed reflects the desires of the RAID community as well as the
network management community.
Information on RAID systems
The RAID Advisory Board (RAB) is the standards body that determines what
is a disk array. See the book The RAIDBook for definitions of the
various terms and level involved with arrays. The ISBN for the fourth
edition is 1-879936-90-9. A fifth edition should be available by early
in 1996. The RAIDBook is periodically updated to reflect advances in
storage array technology.
In addition, a storage array model is defined by ANSI X3T10 Project
1047D, the SCSI-3 Controller Command (SCC) definition. This document is
an ANSI Draft Proposed Standard. No significant technical changes are
anticipated prior to its approval. Revision 6 of the document, the most
recent draft, may be obtained at:
ftp://ftp.symbios.com/pub/standards/io/scsi3/scc-r06.ps
(PostScript format). The SCC model is the basis of much of the RAB model
definition.