Current Meeting Report

2.1.16 Network Data Management Protocol (ndmp) Bof

Current Meeting Report

Minutes from the IETF NDMP BOF held 12/11/01 in Salt Lake City

Approximately 70 people attended the meeting.

Greg Linn presided over the BOF. I have asked Greg to post his presentation to the reflector.
Throughout the BOF, questions were asked and comments were made by numerous audience members.

The agenda, history, and charter were presented.

When asked, only about 25% of the audience indicated they had read the NDMP spec.

During the presentation of the charter a question was asked regarding the willingness to substantially change the existing NDMP standard, if necessary, to incorporate IETF input for the purpose, for example, of improving security.
The answer given was that the NDMP workgroup was very open to making such changes in V5.

The following top 5 V5 requirements were presented as outlined in the V5 requirements document that has been posted to the email list and on the site.

* Multi-source/multi-destination endpoints. Presented by Jim Ward
This will allow multiple slow data sources to be sent to a single tape device and allow a single fast data source to be split between multiple tape devices.
A translate object will perform this multplexing function.
Translate objects could also be defined to perform functions such as encryption.

* Restartability and checkpoints. Presented by Jim Ward
It was noted that this requirement is complex and made more so by the architecture changes likely to be made to support translate objects.

* Security. Presented by Clive Hendrie
V5 needs to address stronger authentication methods, data integrity, privacy (encryption), port range control to better deal with firewalls, and NAT issues.
Questions were raised about what is currently supported for authentication and privacy.
Questions were raised about redundancy with other IETF areas. The NDMP workgroup needs to research this.

* Snapshots. Presented by Harald Skardal
Two types of snaphots - persistent and temporary.
There was a question about the relationship of the protocol to the snaphot format.
It was explained that NDMP will just be used to control the snapshot but not specify the format.

* Management of data objects beyond file systems. Presented by Eric Kadison
Example objects:

* partial file backups of just the part of the file that changed

* databases

* composite files

There was a question about how NDMP will deal with arbitrary file meta data.
This needs to be researched.

There was a question about where backup in NDMP relates to the disk/filesystem/application model.
This digressed into several long comments and questions stemming from a general misunderstanding of the concept of an opaque data stream and the fact that the format of the data passed over the data stream is outside the scope of the protocol.

One point that did come out of this was the necessity of the work group to consider what parts of NDMP are outside the scope of the IETF. This has to do with separating backup domain specific issues from protocol issues.

Greg presented the proposed work group deliverables.
A question was asked about the plan for V4.
This is something the work group must decide.
The preference is to publish V4 as a work group informational document.
The suggestion was made to add this to the deliverables.

Greg presented the Milestones

Greg asked for Comments.
There were none.

During the Summary, Greg asked the audience the following questions:

Is NDMP evolution a useful endeavor?
Several people spoke to the effect that it was.
It was pointed out that NDMP can leverage IETF experience is several areas.
The workgroup needs to be structured such that the IETF does not need to become backup experts in order to assist the workgroup.

Is NDMP appropriate as an IETF effort?
A question was asked if the NDMP workgroup understood the implication of giving up control of the spec to the IETF.
Greg answered that the workgroup did.

Should a workgroup be formed?
A majority of the audience members raised their hands.
There were no dissenting votes.

Do we have willing volunteers?
Approximately 20 audience members indicated their willingness.

Next steps.
Refine the V5 requirements.
Refine the charter.

Workgroup positions.
The already agreed upon positions were presented.
The scribe position is still open.

The BOF concluded with a presentation of where NDMP documents and other related information could be obtained.
A suggestion was made to split the ndmp-tech email list into two lists - one for protocol design issues; another for technical and implementation issues. This needs to be discussed by the workgroup.

Additional comments from workgroup:

The presentation did not include an NDMP architectural overview.
This resulted in some basic misunderstandings that required clarification during the meeting. Such an overview should be presented at a future IETF NDMP meeting.

A point was made in the BOF that NDMP's use of the term 'Data Management' was not consistent with the IETF interpretation of data management and may result in confusion. This concern needs to be explored with the Application Area Directors to determine if this is a 'nit' an individual perspective, or an issue that could become a gating factor for NDMP evolution and therefore needs to be addressed.

A question was raised about possible overlap between NDMP and OPES. The NDMP workgroup should investigate OPES to assess the technological relationship.