The landscape: - The company is cable tv providing also Internet services - They intend to use FAS920 in NFS environment (all servers are Linux) - At this very moment they are servicing about 80k users but wanting to be able to service more that 200k - Services for users are to be implemented with UID and GID quotas
Requirements: - They need to be able to retrieve quickly information about current quota usage for each specific UID or GID. - The information will be available via: -- Call Center - estimated queries 12 times per minute -- WEB GUI for users - estimated queries 120 times per minute - The quota information should be retrieved within 2 s and should be not older than 30 s - There is not specific requirements regarding method of getting the result (SNMP/RSH/NFS RPC/HTTP are acceptable)
We discussed with them SNMPwalk and quota report parsing but they cannot accept it due to high and growing number of users. The most convenient way would be short series of SNMP queries for specific UID / GID.
Any suggestions other than "make and parse quota report" are welcome.
Best regards,
Jacek
Le 21/02/2005 à 10:35, Jacek Nogala a écrit:
Hi,
The landscape:
- The company is cable tv providing also Internet services
- They intend to use FAS920 in NFS environment (all servers are Linux)
- At this very moment they are servicing about 80k users but wanting
to be able to service more that 200k
- Services for users are to be implemented with UID and GID quotas
For such a high amount of users I would not recommend using uids/gids and use virtual users where possible but that depends of the services you are using your netapp for (email ? web hosting ?). Virtual users usually offer much more flexibility. (that's a long topic but to make it short it makes your architecture more robust in case your authentication server goes down).
The same idea applies for the quotas. Netapp's quotas are great but they have a significant performance impact. Depending on the services you are offering, you should use virtual quotas (quotas handled directly by the application which is accessing the data : smtp server, ftp server, ...)
- The quota information should be retrieved within 2 s and should be
not older than 30 s
That's quite a strong requirement, what kind of data is concerned here ?
If this only email or web hosting, using virtual quotas you will get the quota usage by reading the file in which the application stores it (maildirsize for Maildir storage for example). The only drawback of this approach is that you have to provide some access to the nfs data to the management app.
janosldx@gazeta.pl (Jacek Nogala) writes:
...
- They intend to use FAS920 in NFS environment (all servers are Linux)
...
- Services for users are to be implemented with UID and GID quotas
...
We discussed with them SNMPwalk and quota report parsing but they cannot accept it due to high and growing number of users. The most convenient way would be short series of SNMP queries for specific UID / GID.
Usage against per-uid quotas can be obtained by using the requotad (100011) RPC, which is implemented on NetApp filers. As this is technology incorporated into the quota(1) command on Unix systems, I'm not sure why you are considering anything else.
Per-gid quotas are more difficult, and maybe the SNMP queries are the way to go. But I have never found per-gid quotas useful in a Unix context: "group filespaces" are better implemented as qtrees (unless you have too many of them). For usage against qtree quotas you can use the "statfs" calls in the various NFS versions, again implemented via standard Unix commands such as df(1).
Chris Thompson wrote:
janosldx@gazeta.pl (Jacek Nogala) writes:
...
- They intend to use FAS920 in NFS environment (all servers are Linux)
...
- Services for users are to be implemented with UID and GID quotas
...
We discussed with them SNMPwalk and quota report parsing but they cannot accept it due to high and growing number of users. The most convenient way would be short series of SNMP queries for specific UID / GID.
Usage against per-uid quotas can be obtained by using the requotad (100011) RPC, which is implemented on NetApp filers. As this is technology incorporated into the quota(1) command on Unix systems, I'm not sure why you are considering anything else.
First of all we are looking for uniform solution for both UID and GID quotas. Second - I have no idea how fast it is. I have tested it in a small scale and looks promising but in the same environment quota report also runs fast...
Per-gid quotas are more difficult, and maybe the SNMP queries are the way to go. But I have never found per-gid quotas useful in a Unix context: "group filespaces" are better implemented as qtrees (unless you have too many of them).
Unfortunately this is the way they chose to go (my customer has a working solution for other NFS storage - we would like to adapt NetApp to this scenario). GID quotas is also several tens of thousands (less but comparable to UID quotas).
Best regards,
Jacek