rmustacc: zfs doesn't have lfs's "segment cleaner" which would empty out a metaslab for fresh sequential writing.

alanc: reads mainly like "whoever ported zfs to linux didn't understand linux's user namespaces"

jbk: I think you have the metaslab spacemap thing (which indeed uses less space to keep track of free/allocated when the slab is nearly full) confused with intent logs (which use blocks that aren't marked allocated in the spacemaps..)

yeah it could be the spacemaps i'm think of... it's been a _long_ time since I've had to look that deeply :)

intent logs seem like they'd work well since you're really just appending

in other news I gave Hans's NVME host memory buffer patch (code.illumos.org/c/illumos-gate/+/4929) a spin on a test machine where it would used and .. I guess it works? nothing blew up, things seem to be performant.

→ CODE REVIEW 4929: 18307 nvme: want support for Host Memory Buffer feature (NEW) | illumos.org/issues/18307

I don't know nvme-level things well enough to seriously review it.

jbk: until you start reusing the intent log blocks when the transactions commit, and then you're scraping shingles off the roof and reshingling..

sommerfeld: That's good news. How much memory did the controller want?

I started to work on #18324. I would initially disable the superfluous code with #ifdef, or should I remove it immediately?

I'd just remove the code to create the old NOTE segment entirely.

illumos#18324

hrm..

it's a debug build, but bge seems to periodically freeze for a few secs at a time...

curious if anyone else has seen this...

I haven't noticed this personally. What bge part? 5719/5720?

5720

it's possible it might be a victim, but i've got the iperf running on the ice nic on the same box, and it's going fine when those pauses happen

i've not really looked at the driver much (yet)..

rmustacc: nvmeadm get-features reports " Host Memory Buffer Size: 65536KiB"

(it also reports: Temp. Sensor 1 Over Temp. Threshold: 65262C and Temp. Sensor 1 Under Temp. Threshold: -273C which looks suspiciously like 0K to 65535K)

(in this case K=Kelvin)

it identifies itself as a "MSI M450 1TB V1"

per nvmeadm identify-controller, it also has Host Memory Buffer Preferred Size: 65536KiB and Host Memory Buffer Minimum Size: 65536KiB so it's 64MB or go home.

pasted the relevant bits into illumos.org/issues/18307

→ FEATURE 18307: nvme: want support for Host Memory Buffer feature (New) | code.illumos.org/c/illumos-gate/+/4929

[illumos-gate] 18352 i2c ioctl tests race device tree reconfiguration -- Andy Fiddaman <illumos⊙fn>

[illumos-gate] 18327 zen topo module fails enumeration when hardware threads have no logical CPU -- Andy Fiddaman <illumos⊙fn>

[illumos-gate] 18305 fix ktest_get_fn(9F) typedef example -- Robert Mustacchi <rm⊙fo>