|
If you want a stable ABI you use RHEL.
|
# ? Jul 19, 2022 23:20 |
|
|
# ? May 25, 2024 20:30 |
|
RHEL guarantees a stable kernel ABI for drivers?
|
# ? Jul 20, 2022 01:53 |
|
what's a driver?
|
# ? Jul 20, 2022 02:10 |
|
a miserable pile of I/Os
|
# ? Jul 20, 2022 03:20 |
|
an opportunity for incompetency on a grand scale
|
# ? Jul 20, 2022 03:44 |
|
The nice thing about Linux is that the person installing the OS is always doing it wrong.
|
# ? Jul 20, 2022 04:42 |
|
eschaton posted:RHEL guarantees a stable kernel ABI for drivers? The kernel version is locked and only security patches and backported iirc.
|
# ? Jul 20, 2022 04:48 |
|
pseudorandom name posted:an opportunity for incompetency on a grand scale I thought k8s was the solution to scaling Linux (along with an orchestration system atop it) after all, why just use VM, JCL, and the like when you can reinvent it?
|
# ? Jul 20, 2022 04:50 |
|
FlapYoJacks posted:The kernel version is locked and only security patches and backported iirc. does RHEL guarantee this, or a particular version of RHEL? if Linux were a grownup operating system, a driver built for some architecture running RHEL 6 would continue to work without even a recompile on RHEL 8
|
# ? Jul 20, 2022 04:53 |
|
eschaton posted:does RHEL guarantee this, or a particular version of RHEL? Let me know what “grown up” operating system does this.
|
# ? Jul 20, 2022 04:55 |
|
oh man the flexibility of not being tied down to a specific kernel driver binary abi is probably yet another good reason why linux is so widespread
|
# ? Jul 20, 2022 05:50 |
|
having said that, google is working on a binary driver abi for android so your wildest dreams will finally come true
|
# ? Jul 20, 2022 05:52 |
|
*tries to install a windows 2000 driver on windows 11 * Why doesn’t this piece of poo poo work? I thought Windows was a grown-up OS! FlapYoJacks fucked around with this message at 00:51 on Jul 21, 2022 |
# ? Jul 20, 2022 06:12 |
|
So much of the community is scared of recompiling, holy poo poo. Hot take: if you cannot recompile the binary, then don't use the binary.
|
# ? Jul 20, 2022 11:17 |
Athas posted:So much of the community is scared of recompiling, holy poo poo.
|
|
# ? Jul 20, 2022 11:51 |
|
eschaton posted:a miserable pile of I/Os IRQ
|
# ? Jul 20, 2022 12:11 |
|
Splicer posted:But enough talk.
|
# ? Jul 20, 2022 12:21 |
|
eschaton posted:if Linux were a grownup operating system, a driver built for some architecture running RHEL 6 would continue to work without even a recompile on RHEL 8 good bit
|
# ? Jul 20, 2022 14:56 |
|
|
# ? Jul 20, 2022 16:01 |
|
Dracula getting killed by a preemptive scheduler.
|
# ? Jul 20, 2022 16:47 |
|
|
# ? Jul 22, 2022 01:25 |
|
|
# ? Jul 22, 2022 07:56 |
|
Splicer posted:But enough talk.
|
# ? Jul 23, 2022 02:02 |
|
ooh maybe they'll finally tick off that missing feature that zfs zealots keeps nagging about [PATCH RFC v2 00/16] btrfs: add fscrypt integration
|
# ? Jul 24, 2022 22:19 |
looking forward to hearing all the stories of dataloss
|
|
# ? Jul 24, 2022 23:17 |
|
excellent, fantastic idea, next, btrfs should add transparent WAL transaction support for individual files or hey, multiple files at once, why not also let's add support for B-tree files as well so that people can roll their own ISAM implementation easily on top of these features. it's right there in the name, let the user instantiate the filesystem's data structures directly while we're at it let's add an ioctl that performs transparent audio and video compression and decompression, oh and maintains a seek index as well obviously
|
# ? Jul 24, 2022 23:38 |
|
lol
|
# ? Jul 25, 2022 01:53 |
|
Sapozhnik posted:while we're at it let's add an ioctl that performs transparent audio and video compression and decompression, oh and maintains a seek index as well obviously I don't know what the rest means but did anyone ever try to make thing a thing???
|
# ? Jul 25, 2022 02:03 |
|
We built a system for archival of large files. It is extremely slow "all of a sudden" and I was called in to investigate. It is completely swamped with random iops, and just an utter deluge of stat() and whatnot calls, just absolutely murdered. Here's a file size histogram of just one directory. pre:1k: 669933 2k: 119527 4k: 125188 8k: 115541 16k: 97343 32k: 75153 64k: 52402 128k: 29671 256k: 11723 512k: 3304 1M: 755 2M: 42
|
# ? Jul 25, 2022 15:49 |
|
Antigravitas posted:We built a system for archival of large files. sounds like its not working to archive large files
|
# ? Jul 25, 2022 15:51 |
|
That's a conclusion I have reluctantly come to as well. It was built for nice, clean, large streaming writes and reads to saturate 10Gbps. Those were the requirements. Instead, someone is using it in literally the absolute worst access pattern possible. ngl, I feel a bit betrayed. We were assured it wasn't going to be used in this way. MATLAB
|
# ? Jul 25, 2022 16:20 |
|
Antigravitas posted:We were assured it wasn't going to be used in this way. were you assured in writing? if you were, crack open a cold one and relax with the clean conscience of a job well done if you weren't, now you know why you should have been
|
# ? Jul 25, 2022 16:40 |
|
The stakes aren't that high, but yes, I have the documentation that they specifically asked for the opposite of this. But regardless, no file system will perform well when used like this. Lmao, not all file system metadata fits in cache and any directory walk generates a storm of random I/O. I'll have to actually help them get their code to not do the thing they are doing…
|
# ? Jul 25, 2022 16:58 |
|
seems like reiserfs would kill it in this situation
|
# ? Jul 25, 2022 16:59 |
|
just throw some optane drives at
|
# ? Jul 25, 2022 17:02 |
|
Perplx posted:just throw some optane drives at I do not want to enter the university finance labyrinth. Not my budget . Sure, special vdev, l2arc, more RAM. Is there still money left in the grant? Who knows, not me. Putting lipstick on a pig anyway, the system is designed for the opposite use pattern and it's working as designed. It's either educating users to use it as designed, or design another system. That's what I've communicated to the work group.
|
# ? Jul 25, 2022 17:40 |
Optimizing for that workload with allocation classes and special vdevs can absolutely be done, but not on an existing pool that's made for handling large files at 1Gbps linespeed. I don't envy you, it's a poo poo situation with no good outcome.
|
|
# ? Jul 25, 2022 18:06 |
|
or maybe just dont use zfs for anything
|
# ? Jul 25, 2022 19:23 |
|
Antigravitas posted:MATLAB
|
# ? Jul 25, 2022 20:26 |
|
|
# ? May 25, 2024 20:30 |
|
linus did his usual sunday release, which isn't surprising, but...quote:On a personal note, the most interesting part here is that I did the
|
# ? Aug 1, 2022 00:14 |