Messages in this thread | | | From | Manfred Spraul <> | Subject | [PATCH 0/3] ipc: Further updates to sysv/mqueue limits | Date | Thu, 29 May 2014 20:46:37 +0200 |
| |
Hi all,
a) If we increase SHMMAX/SHMALL, then it makes sense to increase MSGMNI, too. And: This allows to remove the automatic scaling (~300 lines)
b) We can also increase SEMMSL, SEMMNI and SEMOPM
c) I think it would make more sense if a namespace starts with the limits from it's parent: If an admin set the limits, then he probably wants that these limits also apply for a new child namespace.
All patches are RFC - they compile, but that's it.
TODO: - check if the sysv sem limits are sane. Especially the SEMOPM - if real users exist that pass > 1k ops, then switch from kmalloc to vmalloc.
@the Redhat developers: Do you have any idea where this "often recommended" comes from? https://access.redhat.com/site/documentation/en-US/Red_Hat_Enterprise_Linux/5/html/Tuning_and_Optimizing_Red_Hat_Enterprise_Linux_for_Oracle_9i_and_10g_Databases/sect-Oracle_9i_and_10g_Tuning_Guide-Setting_Semaphores-The_SEMOPM_Parameter.html
- copy Davidlohrs explanation for the sysv shm limits to sysv msg and sysv sem.
- check if we should also increase the limits for posix mqueue
- decide if it would make sense to increase IPCMNI:
Right now, it is 32768. This means that after 65536 pairs of semget()/semctl(IPC_RMID), semget() will return the same identifier again - and a semop(old_id) won't return -EINVAL, instead it will access the "new" array, which is probably now what the caller intended to do.
The split is arbitrary - we could also split it 1048576/2048 or any other split we want.
- test everything.
-- Manfred
| |