Flush ARC without reboot?

[GeoffK]

I have a problem with a server thats in need of an update.

Its SMB/NFS/iSCSI host backing 5 VM's and about 3TB of shares. iSCSI is there to keep MS happy with SQL/Exchange DB's (no thin provision) and the rest is there either running VM's or hosting files used by ~30 users.

Server is an e3 Xeon w/ 32GB of ram - and at around 35-37 days she runs out of either Arc or Arc Meta (judging from the last ZFS box i had these issues with - it was the meta).

Now this is going to be replaced (e5 w/ 64GB of ram - scope to expand if we need) - but in the mean time - short of rebooting the world - is there anything I can do to flush the ARC and bring storage latency down to sane levels (1-2ms) from sheer lunacy (11-20 seconds).

Thanks.

 

[dlavigne]

Were you able to find the answer to this (or sort it out)?

 

[anodos]

I have a problem with a server thats in need of an update.

Its SMB/NFS/iSCSI host backing 5 VM's and about 3TB of shares. iSCSI is there to keep MS happy with SQL/Exchange DB's (no thin provision) and the rest is there either running VM's or hosting files used by ~30 users.

Server is an e3 Xeon w/ 32GB of ram - and at around 35-37 days she runs out of either Arc or Arc Meta (judging from the last ZFS box i had these issues with - it was the meta).

Now this is going to be replaced (e5 w/ 64GB of ram - scope to expand if we need) - but in the mean time - short of rebooting the world - is there anything I can do to flush the ARC and bring storage latency down to sane levels (1-2ms) from sheer lunacy (11-20 seconds).

Thanks.

Caveat - not a zfs guru. If the problem is truly size of arc meta, couldn't you increase vfs.zfs.arc_meta_limit from 25% to 50% as a stop-gap till you get new hardware.

https://www.freebsd.org/doc/handbook/zfs-advanced.html

 

[jgreco]

Is this some sort of dedup issue, or can you further explain on why you think ARC meta space is having an impact on your performance?