Sysadmin Mega Thread

[oO.o]

I have an old megaraid in a windows machine and the behavior is so weird, I hate it.

[SgtAwesomesauce]

[oO.o]

https://packages.ubuntu.com/source/bionic/zfs-linux
https://packages.ubuntu.com/source/focal/zfs-linux

[SgtAwesomesauce]

is on 19.10 until there’s time to rebuild on zfs root 20.04

[oO.o]

What does the number after Ubuntu stand for? On 20.04 it’s 12…

[Dynamic_Gravity]

Interesting, mine is actually ahead of yours. (CentOS 8.1)

[lbrown@hyperion ~]$ sudo zfs --version
zfs-0.8.3-1
zfs-kmod-0.8.3-1
[lbrown@hyperion ~]$ cat /etc/redhat-release 
CentOS Linux release 8.1.1911 (Core)

[Dynamic_Gravity]

I would assume an internal patch level?

[SgtAwesomesauce]

Tends to be release number.

If they change associated config or whatever, they’ll iterate that without needing to iterate the version.

[SgtAwesomesauce]

@freqlabs

The throughput of a bunch of disks in aggregate is pretty good. SLOG helps reduce latency for sync writes, if you have a very low latency SLOG device and it’s not being overworked. Most of the tests shown are async workloads, and latency isn’t even measured.

Right, but I thought that ZFS, when there’s no SLOG, wrote the transactions to the member disks, then flushed the transaction into the long-term storage format, effectively doubling the write jobs? Do I have that wrong?

[freqlabs]

You’re writing the ZIL on sync and then replaying it to the disks. Without a SLOG the ZIL is on the disks. With a SLOG the ZIL is on a separate device. So either way you’re writing twice. With sync workloads, latency is paramount because every IO has to wait for the previous IO, big or small. You do offload the extra ZIL throughput for sync writes when using a SLOG, but it’s not going to give you more overall throughput if the SLOG can’t keep up. It can be a bottleneck if its throughput doesn’t match the array. As always, “it depends”

[oO.o]

Wow, native Windows package manager

News Story Dump Thread - Stories Only

[TheCakeIsNaOH]

Finally.

Seems a lot less flexible then Chocolatey or Scoop as a package is just a yaml file that has metadata, and software is always downloaded at runtime. Both of the others can have a powershell script run, which opens up way more options. Actually, this seems basically like appget.

---

Edit- OH YES; they are planning to support windows store app installation. source

[SgtAwesomesauce]

@freqlabs

You do offload the extra ZIL throughput for sync writes when using a SLOG, but it’s not going to give you more overall throughput if the SLOG can’t keep up.

That’s exactly what I thought. I’m not saying there’s no reason you wouldn’t go with a SLOG, just thought it was curious that ZFS beat mdadm in most cases measuring throughput only even with the extra writes.

[SgtAwesomesauce]

Nice!

I’m psyched.

Now to make it do terrible things.

[freqlabs]

Goes to show just how poorly mdraid utilizes the hardware I guess.

[SgtAwesomesauce]

It would have to be.

I always thought it was pretty okayish at simple stuff

It was more me being dumbfounded by mdraid than impressed with ZFS. I live in a constant state of being impressed by ZFS.

[freqlabs]

Looks like I’ve got a hot date with some firmware update utilities tonight. Everything is broken lol

[SgtAwesomesauce]

Nice. I think I’m gonna rice my old laptop. I’m missing i3.

[Dynamic_Gravity]

Im taken my a day off nerd shit and drinking wine and watching YouTube.

[oO.o]

@SgtAwesomesauce

Now to make it do terrible things.

Every command installs candy crush