ZFS 5TB SMR 2.5 drives + SSD Write Cache?

scline

Member
Apr 7, 2016
62
19
8
32
I have a 4x5TB 2.5 drives (2 Mirrors in a stripe for Raid 10), I purchased knowing the potential performance issues and plan to only really use them as cold storage. This ZFS pool is only for an off-site backup, primarily write and not reads.

Since I had a few 100G SSD SAS drives (high write endurance) lying around, I was thinking about adding one or both as ZLOG to the pool. I am also forcing log-syncs so regardless of protocol (SMB/NFS etc) they are used for writes.

I now see all writes appearing on the SSD's for this pool, and wanted a gut-check. Is this a good way to potentially offset write issues in the future on this SMR pool or is the SSD write-cache thing not even worth exploring due to the workload.
 

gea

Well-Known Member
Dec 31, 2010
2,485
837
113
DE
There is no such thing as a SSD write cache in ZFS. Write cache is always and only RAM (5s of writes in Solaris and native ZFS, 10% RAM/max 4 GB on Open-ZFS). The Slog is not a write cache, it is a protection device of the rambased writecache what means that every write is going to pool over the writecache as a large/fast sequential write and additionally per every write commit to the Slog.

Even with the fastest Slog on earth, sync write is always much slower than unsync writes. You use an Slog for data security (no dataloss on committed writes due a crash) to avoid the extreme performance degration that sync write normally has.

Best bet is: Do not use SMR in a Raid, sell the SMR disks or use them for backup or for archival data (write once then only read).
 
Last edited:

msg7086

Active Member
May 2, 2017
229
60
28
32
ZLOG is an extra step to slow you down. You still do ALL the work that you had to do without a ZLOG, plus, the extra writes on ZLOG.

And if you want to use them as a cold backup, using them as single disks is probably a better option. Dump data in, sequentially, and keep it there until the day for being overwritten, where you delete all the files and dump data in again.

You are very brave to use SMR drives for writing. I have a few Seagate 8TB SMRs. The slowest speed I saw on those drives is 700KB/s. Good luck with your project.
 
  • Like
Reactions: scline

scline

Member
Apr 7, 2016
62
19
8
32
Thank you both for such valued insight. I have always been confused on the ZLOG functionality and your explanations make so much more sense then the online tutorials about them I have read in the past.

I think I will remove all forms of raid (standalone drive in ZFS) and simply rsync all data between 2 drives to give me a "fake" raid 1.

Since this is a copy of a copy anyway, that rsync is more for not needing to sync several TB's of data over the internet/VPN when they eventually fail.