Large numbers of quotas

[jerry]

I'm in a situation where I have a structure that looks like this:

/ifs/user/a

  abc

  albert

  another

  ...

/ifs/user/b

  bob

  betty

  ....

I have about 10,000 such user directories, but they are split a-z (for automounter and other complicated reasons).

I need to keep track of usage on each directory, but I'm torn between:

1. setting a wildcard user quota (if this can be done) that says all users get X gb.

2. setting a quota per directory (this could be pretty crazy)

3. setting a quota only on the "a" "b" "c" etc dirs and using another utility to breakdown further at the time I have a quota alert (like durep).

I'm concerned about having 10,000 user dirs and how to implement it with the least impact, even if it means I have to use some other tools to dig further when a quota full scenario occurs.

Any thoughts would be interesting.

Jerry

[J. Lasser]

I'd set up a wildcard user quota. When a user legitimately requires
more space, you can decouple that specific user's quota from the
wildcard quota and give them more space.

When you say that you're concerned about impact, what specific impact
do you anticipate?

--
Jon Lasser                     j...@lasser.org                      206-326-0614
. . . and the walls became the world all around . . .  (Maurice Sendak)

[Jerry Uanino]

If my cluster is tracking 10,000 users..... does it add any significant overhead?  I'm not sure how the underlying system is effected by every tracking quota.

For example, if I use a tracking quota vs an enforcement quota is that better? 

If you think it's fairly lightweight, then I could prob just do the wildcard and let it rip and see what happens.  I'm also interested if there is any way for me to see the impact on the cluster by any additional overhead I've added with the quotas.

[J. Lasser]

The amount of overhead quotas add isn't significantly impacted by the
number of quotas: you still need to update the accounting domain,
it's just a different accounting domain.

In my experience, it's pretty lightweight.

[Daniel Cornel]

Do you have any advice on how to accomplish this? Im able to create quotas, hard, soft, advisory, via CLI, however I don't know how I would apply them with wildcards.  Can you explain? I have about 3000 users in a directory where this would be handy.

[Peter Serocka]

Create:

isi quota create --path=/ifs/data/xyz --default-user

Lists as:

default-user        /ifs/data/xyz              no   N/A   N/A   N/A  ###G

* user:joe          /ifs/data/xyz              no   N/A   N/A   N/A  ###G 

* user:jane         /ifs/data/xyz              no   N/A   N/A   N/A  ###G 

Therefore "wildcard", but no wildcard character is actually involved

when creating with --default-user.

Specific threshold setting for one user:

isi quota modify --path=/ifs/data/xyz --specific-user=joe --unlink-from-default --hard-threshold=1T

default-user        /ifs/data/xyz              no   N/A   N/A   N/A  ###G

user:joe            /ifs/data/xyz              no  1.0T   N/A   N/A  ###G 

* user:jane         /ifs/data/xyz              no   N/A   N/A   N/A  ###G 

Note that after "--specific-user=joe --unlink-from-default" the "*"

for joe has gone,

so this quota is not "wildcarded" = "linked" to default any more.

There is no advanced wildcarding in the sense of pattern matching,

like --specific-user="student*", or per-user accounting for

a certain group of accounts.

(The result of --specific-user="a*" seems pretty useless on 6.5,

one quota is created just for the very first account that starts with an a.)

-- Peter

--
You received this message because you are subscribed to the Google Groups "Isilon Technical User Group" group.
To unsubscribe from this group and stop receiving emails from it, send an email to isilon-user-gr...@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.

Peter Serocka

CAS-MPG Partner Institute for Computational Biology (PICB)

Shanghai Institutes for Biological Sciences (SIBS)

Chinese Academy of Sciences (CAS)

320 Yue Yang Rd, Shanghai 200031, China

pser...@picb.ac.cn