[Search for users] [Overall Top Noters] [List of all Conferences] [Download this site]

Conference ssag::ask_ssag

Title:Ask the Storage Architecture Group
Notice:Check out our web page at http://www-starch.shr.dec.com
Moderator:SSAG::TERZAN
Created:Wed Oct 15 1986
Last Modified:Fri Jun 06 1997
Last Successful Update:Fri Jun 06 1997
Number of topics:6756
Total number of notes:25276

6558.0. "HSZ40 RAID5 expansion question" by ODIXIE::RREEVES () Mon Apr 07 1997 11:53

    
    My customer has an Alpha Windows-NT cluster system with a HSZ40.
    They have a RAID5 array configured on the HSZ40. 
    
    If they add an additional member to the current RAID5 array do they 
    have to recreate the array or is there a way to add a new member 
    and havethe array expand without performing a complete backup and restore
    operation. 
    
    I believe that they have to perform a backup and restore but I wanted to
    check here to be sure.
T.RTitleUserPersonal
Name
DateLines
6558.1SSDEVO::ROLLOWDr. File System's Home for Wayward Inodes.Mon Apr 07 1997 12:422
	They will have to re-create the array, with the attendant
	backup and restore.
6558.2AMCFAC::RABAHYdtn 471-5160, outside 1-810-347-5160Mon Apr 07 1997 13:4924
How sad.  I wonder when storage engineering will figure this out?  I don't
suppose a temporary RAID 1+5 would work -- can the members of a RAID 1 set be of
different sizes?  Of course this would require basically twice the number of
disks until the copy is done.  An in-place dynamic reliable reconfiguration is
not available yet.

I have thought up an algorithm to do it but haven't tried implementing it yet.

	N = 1
	loop:
		read LBN range N using old mapping
		write LBN range N using new mapping
		N = N + 1
	until done

This glosses over some rather sticky details.  Foremost, the file system must be
enhanced to accomodate the change.  Application I/O must be integrated properly
-- a fence dividing the old portion from the new.  One problem comes from the
write potentially overwriting data before it is read.  Another comes from the
risk of the system failing during the loop.  Finally, performance must be
addressed.

Btw, if the application is layered upon RTR then there does exist a procedure
for expanding the data storage dynamically -- details available upon request.
6558.3Thanks for the answerODIXIE::RREEVESTue Apr 08 1997 17:031
    Thanks for the answer in .1