[OmniOS-discuss] Heavy NFS load causing OmniOS server to become unresponsive (kernel memory)

[Doug Hughes]

I didn't see this show up on the list, so I'm trying again.

----

Been seeing this a bit lately on a server with 96GB ram where the zil is 
limited to 48GB. Under heavy NFS load (caused by user running parallel 
find/xargs/rm clean job), it sends the machine into desparation memory 
and causes it to be unreachable for a while. (the server is 24 x SSD, 
and internal I/O load is ok, but the many concurrent NFS requests cause 
it to die, sometimes requiring a reset. usually it recovers in a few 
minutes, but it becomes unservicible from memory shortage.

I have some mdb/kmastat/kmausers/memstat output.

here's the kmastat break down.

Total [hat_memload] 15.8M    5155240     0
Total [kmem_msb] 22.0G  987112603     0
Total [kmem_firewall] 271M  411704261   895
Total [kmem_va] 40.8G   22314937     0
Total [kmem_default] 31.8G 4275706645   619
Total [kmem_io_4P] 44.7M   26714639     0
Total [kmem_io_4G] 108K        908     0
Total [kmem_io_2G] 100K         68     0
Total [bp_map]                                                0 7251     0
Total [umem_np]                                               0 728     0
Total [zfs_file_data] 118M      93380     0
Total [zfs_file_data_buf] 19.4G   22030035     0
Total [segkp] 256K       6787     0
Total [ip_minor_arena_sa] 512      72304     0
Total [ip_minor_arena_la]                                    64 29350     0
Total [spdsock]                                               0 1     0
Total [namefs_inodes]                                        64 27     0
------------------------------ ----- --------- --------- ------ 
---------- -----

the kmausers is almost 1MB.


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://omniosce.org/ml-archive/attachments/20160422/4cae3ab6/attachment.html>