SLOG with iSCSI SSD pool?

[tortue]

I've got a question I haven't really been able to run down. Everything I've read about using a SLOG for assisting with iSCSI combined with sync=always writes are usually regarding largish pools and thus presumably using spinning disks. My question is, would a SLOG still be recommended when the pool is already an SSD? I figured I'd be ok with a low use system using iSCSI on an SSD with no SLOG, but maybe not.

My question stems from a crashing issue I'm having with a pool shared via iSCSI when enabling sync=always. I'll start another thread on that issue once I reproduce it with less steps (right now I can reproduce it with a handful of actions but I think it can be narrowed down). I don't really expect FN to crash due to lack of a SLOG on a sync=always iSCSI share but if it's due to not using best practices for SLOG/iSCSI then I guess I could accept that and not bother troubleshooting.

 

[bigphil]

Even with an SSD pool, if you want to use sync=always, it's still a good idea to have a separate log device regardless of pool size for at least two reasons I can think of: a SLOG should have power loss protection, very low latency, and write optimized. Your pool drives may not necessarily meet these requirements. There was another recent thread regarding iSCSI crashing on a new build of FreeNAS unless a SLOG was added to the pool...could be an issue.

 

[bigphil]

Here is the thread I was referring to. Is your iSCSI initiator a Windows box by chance? Just wondering as two users in that thread are also running Windows...not sure on the OP's setup.

 

[tortue]

Here is the thread I was referring to. Is your iSCSI initiator a Windows box by chance? Just wondering as two users in that thread are also running Windows...not sure on the OP's setup.

My initiator is a patched up ESXi 6.5. Thanks for the link, the symptoms look very similar. Though I'm sending a very small amount of data (measured in Mbytes) and I don't believe my RAM was fully utilized except for the first crash as some of my troubleshooting testing was done shortly after a reboot with little usage (so presumably the ARC had not filled up yet). I hope to have the time this weekend to go through and reproduce it while taking notes and troubleshooting it.