home *** CD-ROM | disk | FTP | other *** search
- Xref: sparky comp.sys.sun.admin:9435 comp.unix.admin:6641
- Path: sparky!uunet!dove!cam!fs3!klm
- From: klm@nist.gov (Ken_Manheimer_x3539)
- Newsgroups: comp.sys.sun.admin,comp.unix.admin
- Subject: Group-oriented disk-space apportionment schemes??
- Message-ID: <KLM.92Dec14175003@coil.nist.gov>
- Date: 14 Dec 92 22:50:03 GMT
- Sender: news@cam.nist.gov
- Distribution: comp
- Organization: National Institute of Standards and Technology
- Lines: 85
-
- I'm looking for suggestions and experiences managing NFS disk
- apportionment across large user bases according to their memberships
- in organizational groups.
-
- I'm designing a shared-filesystem service intended for wide sharing
- within my organization. The primary aim of the service is to provide
- the capability for file sharing to those organizational divisions so
- small or non-technically oriented ("technically disoriented"?-) that
- they do not have their own facilities to provide this capability.
-
- The amount of disk space is by no means large relative to the size of
- the prospective clientele, so we will have to be vigilant about disk
- allocations. I envision having to have some nominal allocation, and
- then bump people up as they need more. This individual-account based
- approach has a number of what i see as serious drawbacks. They
- include, off the top of my head:
-
- - Large groups with lots of members can claim much larger portions,
- but large groups are the ones most likely to not have great *need*
- for the service. They are more likely to have their own disk
- service resources, and so already have the capability for disk
- sharing which is the real aim of this service. (Everyone always
- could use *more* disk, but that's not what the service is for. The
- intent is to provide something to the have-nots that they need, not
- supplement the disk space contingent of anyone who wants it.)
- - Usage accounting is almost inevitably most useful when done
- according to groups. This is difficult, with the groups
- allocations scattered all over the map, and the propriety of a file
- sometimes (probably much more than occasionally) being the owner's
- secondary groups, not their primary one.
- - In general, individual-account based allocation means large
- databases that are difficult if not unmanageable for mapping to
- collective allocations.
-
- I would *really* like to be able to allocate disk space portions to
- collective groups (possibly but not necessarily the formal Unix groups
- dictated by the /etc/group file), and have the space available to an
- individual depend on some group of the hierarchy. I do not see a
- manageable way to do this, however. The server will be a Sun machine
- running SunOS 4.1.3. Quotas under SunOS (and any typical Unix OS, as
- far as i know) are purely account oriented,
-
- I've thought about (ab)using other mechanisms, like disk partitioning,
- to divvy out space to collective groups, but
-
- (1) nothing i've come up with would provide the number of units i'd
- need,
- (2) even if some alternate could be finagled to arrange for enough
- portions to go around, there are some real problems with such
- mechanisms when it comes to boundary conditions (eg, exhaustion
- of the space allocated to a group) and reallocation.
-
- As i said, i've considered using disk partitioning to make the group
- apportionments. Nominal individual accounts would be used in
- conjunction with the partitions to control the way the disk space is
- consumed. However,
-
- - (1) Disk partitioning only enables dividing a disk into 7 portions,
- as far as i know - [a-h], minus the 'c' partition, which must
- encompass the whole disk.
- - (2) The only way i can think of to expand the space available to a
- group is by creating moving all their stuff to a new, bigger
- partition elsewhere (or archiving their stuff, recreating their
- partition with more space, and then replacing their stuff in
- the larger space - not a fun process, considering the
- drastically large chances for catastrophic screwups).
- - (2) Users discover their group has run out of space when the disk
- is (effectively) full, and the write fails. (In some
- situations the file would be left partially written, i think.
- Yick.)
-
- What the heck do other people do in this situation? I wonder whether
- anyone has developed a preprocessor produces the quota data files
- according to higher-level group, disk-layout, and defaulting
- information. Is there some approach i'm missing, or some prominent,
- alternate substrate (a mythical NFS++, or the commercial form of AFS),
- that takes on this issue directly and successfully??
-
- Please respond to me via email - i find it almost impossible to keep
- up with these newsgroups - and i'll summarize what i uncover.
-
- Ken Manheimer klm@cam.nist.gov Computer Systems and Computation Division
- W: 301 975-3539 Nat'l Institute of Standards and Technology
- H: 301 762-4868 Technology 225/A151
- Gaithersburg, MD 20899
-