Wanted: Step by Step to add SLOG by GUI and have it wear well

[GeekGoneOld]

I have been running a FreeNAS system with 6 disks as a RaidZ2 config for years. I upgraded to 9.3 STABLE yesterday with complete success. Then I added two S3500 SSDs to (finally) add a slog to the volume.

I have read a thousand posts and the current (and past) manuals but STILL have a few questions.

1.
Am I correct that the basic approach is to go to Volume Manager then
-choose volume to extend as my volume
-click on the + beside the two ssd
-choose Log (ZIL) in the first of two pulldowns in volume layout
-??? do what with the second pulldown???
-click on Add Extra Device

2.
Why would I choose this over the "manual setup" choice which seems (surprisingly) easier:
-click on manual setup
-choose volume to extend as my volume
-in the ZFS extra table, click on "Log" for each of the two SSDs

Finally, I know I don't need anywhere near the whole disk (MAX use would be network actitvity for 10s) so a slog of 1MB would do. That leads to the last question.

3.
Either method above will allocate the entire disk size (80GB) as mirrored slog. Will this compromise wear balancing of the SSD or should I somehow make sure the system only configures a slog of 1MB (e.g. via CLI)? I am comfortable with CLI but always prefer to configure with GUI.

I REALLY hate to ask, but try as I might, I haven't seen convincing answers to these three questions and I this install to go well (even though I have a backup, of course).

 

[dlavigne]

1. Yes.

2. This screen is to provide unrecommended backwards compatibility for users who want to use different sized disks to create a pool. It does not apply to what you want to do.

3. Unless you are running an ancient version of ZFS (which 9.3 is not), a mirrored SLOG is overkill. If an SSD dies, you would only lose the last few seconds of writes which had not yet made it to disk. However, the SLOG should be the entire SSD. Personally, I would save one of the SSDs for another purpose, and if I was worried about wasted space on an SSD, use a smaller one.

 

[GeekGoneOld]

1. Yes.

2. This screen is to provide unrecommended backwards compatibility for users who want to use different sized disks to create a pool. It does not apply to what you want to do.

3. Unless you are running an ancient version of ZFS (which 9.3 is not), a mirrored SLOG is overkill. If an SSD dies, you would only lose the last few seconds of writes which had not yet made it to disk. However, the SLOG should be the entire SSD. Personally, I would save one of the SSDs for another purpose, and if I was worried about wasted space on an SSD, use a smaller one.

1. What about the second pulldown in "volume layout"? What do I put in (or select) in that box?

2. Makes sense. Thanks.

3. I know the issue of the old ZFS15 vs ZSF28 and value safety over cost. My issue is with wearout of the SSD. Does the SSD (S3500) only use unallocated space for its wear leveling and garbage collection (suggesting I should only allocate, say, 1GB for slog)? Does FreeBSD (or FreeNAS) have a wear leveling algorithm (like TRIM) so it doesn't matter? Can I benefit from both if both SSD and FreeNAS can wear level?

 

[GeekGoneOld]

Bump


Sent from my iPhone using Tapatalk

 

[Ericloewe]

FreeNAS supports TRIM. Any reasonable SSD will use all filesystem free space for wear leveling/GC.

 

[Nick2253]

Unless you are running an ancient version of ZFS (which 9.3 is not), a mirrored SLOG is overkill. If an SSD dies, you would only lose the last few seconds of writes which had not yet made it to disk. However, the SLOG should be the entire SSD. Personally, I would save one of the SSDs for another purpose, and if I was worried about wasted space on an SSD, use a smaller one.

If you wouldn't mind, could you clarify this statement?

In my mind, losing one write, much less the last few seconds of writes, would be a big deal, making a mirrored SLOG an important thing (far from overkill).

Is there something I'm missing here?

 

[GeekGoneOld]

Pre ZFSv19, losing the slog lost your entire volume! Now you just lose the last few writes, but ONLY if you lose the slog at power down (since it is normally write only). Actually, the power down would have to be uncontrolled (crash, kernel exception...) or FreeNAS would flush cleanly.

As a result, some people say mirror slog is overkill. Personally, I say that if ecc ram is a must, so is mirrored slog.

 

[GeekGoneOld]

@Ericloewe. I had read that TRIM for ZFS was to be supported in FreeBSD 10. I see now that they got it into 9.2(+). The current 9.3 STABLE thus should have TRIM for ZFS and, presumably, for a mirrored slog.

Not quite sure how to verify...

 

[GeekGoneOld]

After offline discussion with rs225,I really didn't trust that TRIM was of any use. As he pointed out, the slog is (mostly) write only and may not use TRIM anyway. Neither trusting TRIM, nor requiring the space, I resized the disks per the following:

1) installed two new 80GB Intel S3500 (came up as da0 and da1)
2) went to shell and typed (to make a huge HPA area on the drives to leave 8GB)
-camcontrol hpa da0 -P -s 15625008
-camcontrol hpa da1 -P -s 15625008
3) rebooted
4) clicked on Storage|Volume Manager then
-set Volume to extend to the volume I wanted to add the slog to
-clicked the + beside the two 8.0GB drives
-in volume layout set the first pulldown to Log (ZIL)
-clicked on Extend Volume

That was all.

Note that I originally tried this with a fakeraid card (Promise FastTrak TX2200) with the disks set to JBOD and it failed (gracefully, thank goodness!!!) due to not properly handling camcontrol (don't know exact message). I then put in an IBM M1015 crossflashed (see my next post) and all went well.

 

[Ericloewe]

Uhmm, TRIM has no bearing on reads. It is precisely write performance (and NAND longevity) that benefits from TRIM.

 

[GeekGoneOld]

My "write only" reference was not considering the alternative to be "read/write" but rather "write/delete" implying slog may not delete. I don't know (haven't viewed the code) if the write consumes all available space or if it overwrites "from the beginning" after each transaction (5 seconds). Since I couldn't verify whether it ever deletes and wasn't sure if it expanded to fill all space, I took the safe approach.

And, still, I don't know if TRIM applies to slog devices since it is not at such a low level to be "on" or "off". Specifically it was supported originally only for UFS and not ZFS, therefore it is not unreasonable to wonder if it is supported for ZFS but not for slog.

Just being particularly cautious. Belt and braces since I lose nothing!

I would, however, like to know the authoritative answer...