any recommendations for [single port] ethernet adapter? should support 1, 2.5, 10 Gbit. I guess, something like intel x500 series maybe?

tsoome_: What interface? Copper or SFP?

I wish you all fast acceleration for the new year 2024.

rmustacc copper

Cheers, getting a bit lost with HDD technologies and implications for ZFS in particular (e.g. that debacle with CCTV oriented disks that were poor choice for random I/O) - which key words are "good" or "bad" nowadays? :)

looking to update my home system

Looking at Western Digital Ultrastar DC HC580, 3.5", 24TB I see both "CMR" and "EAMR" mentioned - are those "okay"? :)

Likewise, pondering a home NAS update :)

yes and stormont and woodstock

oops, wc

tsoome_: I'm using X550T2 for OmniOS on Supermicro Board (Intel X3)

tsoome_: in 10Gbit mode with switch Netgear S512EM-100EUS - a small Office Setup where a Samba Server make file service.

jimklimov: as I understand it, CMR good, SMR bad. C="Conventional" non-overlapping tracks that can be written in any order. S="Shingled" overlapping tracks, where within a drive-defined track group the drive must write sequentially (with much buffering and added latency if the drive firmware tries to hide this from the host). EAMR+CMR should be good.

even if an SMR drive presents it as a conventional block device (as opposed to zone storage), zfs really doesn't deal with it well at all (as much as you'd think it'd actually be a good fit)

SMR comes in two flavors: drive-managed and host-managed. host-managed requires special software to enforce the correct write ordering. drive-managed tries to hide the complexity in hardware but can't hide the latency hit.

jbk: IMHO the missing interface for drive-managed SMR is some sort of indication from the drive of how backlogged it is.

I should have said "hide the complexity in firmware".

IIRC a ZFS resilver on a "dirty" drive is a worst-case workload for SMR. Unending stream of writes, mostly not well-aligned with the SMR zones.

even then, i suspect some of the other bookkeeping bits of zfs just don't play well.. or at least HDDs are just too slow

on linux you can do that 'hide the complexity' at the device layer, and zfs still has lots of problems w/ it

tomww thanks, I was looking for something similar:)

thanks all! happy holidays :)

tsoome_: hm?

jbk: yeah, I suspect the flow control needs to extend upwards into ZFS.

unfortunately, zfs itself doesn't seem equipped for such things

we've hit several problems that end up all being variants of 'zfs has no concept of backpressure'

e.g. prefetching can exhaust all ram if your disk isn't fast enough

Does Sun make (the default of OI/OO/...) have some syntax for exporting make vars as envvars to launched programs? GNU/BSD both support `export VAR=VAL` but not Sun :\

The `man make` page a bit cryptically says that:

make exports environment variables, using the most recently defined value. Macro definitions supplied on the command line are not normally exported, unless the macro is also an environment variable.

make does not export macros defined in the makefile. If an environment variable is set, and a macro with the same name is defined on the command line, make exports its value as defined on the command line.

so I guess if I go either `VAR=VAL make`, or `export VAR ; make VAR=VAL` or `export VAR; make` (and have `VAL=VAR` assigned as a macro in the makefile) then it gets exported further to build toolkit programs - but only then, if the caller shell takes care to export them?

jbk: yep. 13093 covers a few. And (after a few panics during a weekly scrub) I added one that throttles scrub when freemem is low.

jimklimov: while it doesn't directly answer your question, note the -e option to make (which imports environment variables into make)

jimklimov: in the make rule you can set environment variables for the specific command.

test:

FOO=${BAR} printenv FOO

(the usual /bin/sh syntax for per-command environment variables)

yeah, but in my case I want something applied for any use-case of a $(CC) or $(CXX)

technically, they can be (symlinks to) ccache and I want certain CCACHE_* envvars applied whenever a developer re-runs `make` in the configured workspace while iterating, without requiring that stuff is pre-exported in their shell

I can surely pre-export them in a CI wrapper script which does autogen+configure+make dance, but for interactive use it is cumbersome compared to editing a file and writing `make` anywhere relevant in the build tree

you can actually stuff the environment variable settings inside $(CC) or $(CXX) if you're feeling sufficiently twisted.