SLOG Optane DC P4801X 100GB with 512GB of RAM

[octavez]

Hi,

Would there be a write performance increase if adding an M.2 Optane DC P4801X 100GB for SLOG knowing there is already 512GB of DDR4 RAM?

I'm wondering if the SLOG capacity as to be higher than the RAM capacity to improve performances?

Thank you.

 

[jgreco]

There are zero cases where sync writes are faster than async writes. Your write cache is not going to be more than a small fraction of the 512GB of RAM, and a SLOG device only needs to be large enough for a few transaction groups. If you are laboring under some errant idea that a SLOG device is a write cache, please disabuse yourself of the notion. It is only to make writes that are specifically sync writes, faster, because writing to the in-pool ZIL is somewhat slow.

Some insights into SLOG/ZIL with ZFS on FreeNAS

What is the ZIL? POSIX provides a facility for the system or an application to make sure that data requested to be written is actually committed to stable storage: a synchronous write request. Upon completion of a sync write request, the underlying filesystem is supposed to guarantee that a...

www.truenas.com

 

[octavez]

Thanks @jgreco, that's what I was thinking! So in this configuration below you would not recommend SLOG?

Our build is as follow:

AMD EPYC 7402
60 x Seagate Exos 20TB (Vdevs of 10 HDDs in RAIDZ2)
512GB RAM (8 x 64GB DDR4)
L2ARC: 2 x Micron 5300 PRO 3.87TB

It is used for video post-production. 3-4 times a week a 10-20 terabytes of footage are transferred to the server. Then for the rest of the time footage is access by a dozen of workstations at the same time for editing (mostly read).

We have another build pretty much identical with a bit less ram and no L2ARC, that is used as a mirror with replication tasks every night.

 

[HoneyBadger]

That workflow doesn't write synchronously, so an SLOG will be pointless there.

Where are you seeing performance bottlenecks (reads or writes?)

 

[jgreco]

SLOG wouldn't be useful here.

 

[octavez]

Thank you all! Currently building the servers, so haven't done performance tests yet.

Would there be any reason going with SCALE over CORE? I'm thinking if it's easier to manage the mirror server, as sometimes replication tasks can be fairly slow with CORE between two servers. But I've heard that CORE could be slightly better for performance, which is the most important thing for us.

 

[NugentS]

For performace Core > Scale at the moment

 

[kspare]

I found those 4801x 100gb drives to not be very fast. I ended up running 2 P3700 800gb drives instead.