IAM
======

Name:		Data Migration Manager
Submitter:	Mark Cruciani
Owner:		Rick Matthews
Intern:		Dean Roehrich
Interest:	
Status:         inception scheduled 05/21/2008
Exposure:	open
Comment:	


SUMMARY
=======
 An open-source tool which facilitates the movement of file
 systems from one platform to another over the NFS and/or
 CIFS protocol.  Read/write access to file system
 content is maintained over the duration of the migration.

ISSUES
======

gcs-1 Glenn Skinner: Would like to see drive through of use cases.
====>Provide this info in materials.

ANSWER: Other attemps at Sun to consolidate migration tools have been 
complicated. This one is simple.

Typical case would be an administrator, who has permissions.
If admin is migrating a single file system from a diff storage appliance to a 
new file system on a diff platform, user would start up
using GUI controls. Admin could control policy and priority
orders. Migrator would pause and have user point to client on
target. Migrating from one piece of online storage to another piece
 of online storage.

Q: Can you migrate from online storage to offline?
A: It would be from one active read-only Filesystem to another active filesytem.

Q: Are you interfacing with block storage or with files?
A: It's done exclusively with files / named attributes / extended attributes.
Migrates everything on NFS and CIFs client.

Q: Why do you need restriction of Read only on storage side?
A: Upon migration there is no mechanism to check for updates.

General intent was to ...a fully mounted NSF file system...
No direct control of it's IO to having direct control of the IO.

Q: Doing translation of ACL's beteen NFS and GFS?
CIFs service has ability to do mapping.
NFS will move without translation. Need to determine this.
Not to be in business of doing translations.

Q:  A local file sys you are going to migrate to GFS file sys.
A:  The ACL's are diff. Who is going to handle mapping if
translation is diff?

Whatever we are getting from NSF is the restriction of
what we can put back into GFS.
======>Project team to investigate further.


ARC Issues file for PSARC/2007/692

ram-1 How do you intend to set up and administer DMM? Is there any significant
      differences whether in a storage appliance, or on a general purpose
      computer system?

======>The GUI specifies your engine, storage, platform, share and target.
At that point in time DMM creates the target. All is controlled by the GUI.
First screen is setup screen. A resource meter exist. User can specifiy how 
much resource to dedicate to migration and user specifies a migration schedule.

Operations are identical to other migration tools.


ram-2 Is there a scheme for mapping identities in DMM? for CIFS users?
      for NFS users?

======>No plans for local file systems. In Non-local case we are relying
on NFS and CIFS protocols. Currently looking at supporting NFS v3 and v4.


ram-3 Is there any mechanism to prevent DMM from being used as a general
      FS replication service? Does there need to be?

======>Nothing prevents from having mulitple versions/instances of DMM being
fired up. Can use it as a replicator. Can replicate a source to mulitple FS
on a target OR replicate uniqe FS on multiple diff platforms.

ram-4 Other than access to a file eligible for migration in DMM, does an
      end user (excluding administrator) have any other control or
      visibility into the process?

======> Admin has full control. End user should not know the migrator is
in place. That is the goal. They can access the file and read/write to target.

Q: Glenn Skinner: What is the intent of the project?
A: copy, cpio, tar are effective migration strategies.
We are trying to address cases where you do a copy or tar
you lose access to data. You have a terabyte of data on a system. 
You want to upgrade. Applications need data during migration, 
coventional systems  don't have it. Intent is to set up migration 
clients and targets and fire up the migration. Data is read-only on 
source / read-write on target.

Q: What is expected behavior on reboot?
A: Migrator can't lose data.
Using mySQL database which tracks namespace during migration.
Using CIM monitors. A write has come in that haven't been committed 
to ZFS would be lost during crash. 

DMM is hosted on the target platform.
Write case on a section of file that hasn't been modified.
If the source crashes, we lose access to it and it waits until
it comes back..NSF not responding.

Running into issues when rexporting NSF mount.
Target FS The issue is with  ... is outside scope of Monarch at this point.
The target file system 


THE NEXT STEP
=============

-Include the motivation of project in materials for next round of review.
-Investigate further translations of ACLs between NSF and GSF.
-Take up any questions with case owner.