[410] in Moira

home help back first fref pref prev next nref lref last post

Moira backend

daemon@ATHENA.MIT.EDU (Richard Basch)
Sat May 30 15:40:59 1992

Date: Sat, 30 May 92 15:40:23 -0400
To: moiradev@MIT.EDU
From: "Richard Basch" <basch@MIT.EDU>


I had an idea for easier operational management - an additional Moira
table and Moira commands.  Given that it seems that we may have to make
the afs.incr command a Moira client, anyway, we might as well make it
easy for Operations to manage the allocation tables for AFS.  Currently,
it seems that the back-end will be using a table of allocations/usages
which can only be managed via root on Moira.

If we can somehow include it in the Moira interface, that will simplify
Operational management.

Here is a proposal of a revised "physical" table (trying to account for
both NFS and AFS).

   phys_id would correspond to nfsphys_id.
>> type would correspond to filesystem type (NFS,AFS)
   mach_id would correspond to mach_id of the machine.
>> cell_id would correspond to mach_id of the cell (0 for NFS)
   device (same as before)
   dir (same as before); for AFS, it would be something like /vicepa
   status (same as before)
   allocated (same as before)
   size (same as before)
>> usage
   mod... (modtime,modby,modwith)

The problem is that we will have to put the allocation rules into the
Moira server.  Anyway, with the addition of the "usage" field, which
would no doubt be updated from an external source, we could try
something like:

	max [(2*free space) - (unused allocation)]
	= max(2*size - usage - allocated)

For "zero" usage fields, we may have to guess to do a comparison.
Zero usage fields probably indicate that the usage/allocations are not
being adjusted automatically.

	Assume usage ~ x*allocated; the equivalent equation would be:
		max[2*size - (1+x)allocated]

	From initial findings, x ~ 0.65.

Comments?

-Richard

home help back first fref pref prev next nref lref last post