Notices by Chris Siebenmann (cks@mastodon.social)

@ewenmcneill @lanodan We have some servers at work that support four U.2 NVMe drives, basically as 'system' drives, and I know some vendors sell ones that have many more (eg, Supermicro I believe has models that do 16+ U.2 NVMe).

@ewenmcneill @lanodan Four-way passives (requiring an x16 slot with bifurcation support) seem widely available around here in online retailers. Of course your typical motherboard probably doesn't have very many real x16 slots, so you're not going to build a big storage system that way.

My impression is that NVMe-based storage systems use custom server motherboards that break out a lot of PCIe lanes to eg U.2 slots (and I don't know how they handle hotswap, but they do).

@lanodan Since NVMe is PCIe in a different form factor, multi-drive adapter cards are either very simple but require PCIe bifurcation from your BIOS (not necessarily available) or quite expensive because they need a PCIe bridge/switch/etc to multiplex multiple downstream NVMe drives onto one set of PCIe lanes.

(The cheap require-bifurcation ones just split eg x16 PCIe lanes to 4 x4 sets and wire them to four NVMe drive slots.)

This is my face when 2.5" SATA SSDs larger than 4TB appear to be, shall we say, extremely uncommon (and $$$$), while 8 TB NVMe drivers are garden variety and not unreasonably priced. Look, people, I have a lot more SATA ports than spare PCIe slots (even 1x slots, which you can get NVMe adapters for).

Why am I looking at this? Music FLACs, that's why.

@ben AFAIK my university's use of Azure specifically requires our data to reside in Canadian datacenters, operated by the Canadian Azure/Microsoft subsidiary (does that actually happen? not my department, literally). This was at one point considered to be good enough by decision makers who could sign off on that. Now the public commotion over this may cause a change (and that's why said commotion is important to me; the more of it the better).

There's some GNU Emacs hackery I could and should do (to make it properly open links in my environment even when it's not running with $DISPLAY set), but it involves dealing with regular expressions in GNU Emacs and ... well.

I am sure that people who do this a lot do not stare at the level of peculiar and opaque quoting required and back away slowly.

@mos_8502 My understanding is that one initial driver of this was that credit cards were much less common in Japan in the early days of online purchasing. So businesses got ingenious and did deals with convenience store groups and so on.

(I like the result, it has a pleasing degree of separation between various parties.)

@catzilla I'm also kidless and my incidental exposure to those of relatives (and the relative who's a teacher) hasn't covered modern schoolwork. I think at the university most everything is submitted electronically these days and professors/TAs would give you funny looks if you wanted to submit a printed copy, much less a hand-written one.

(This makes the central coursework submission system extremely critical and I'm forever thankful that I'm not one of the people looking after it.)

@catzilla My cynical side suspects that teachers don't want to try to read (current) student handwriting, and then argue with the parents of students when they mark students down because "the handwriting is too hard to read, get better at it".

Also probably they'd have to assign less writing as homework because it takes longer than typing/etc.

(In retrospect, I have no idea how my HS teachers coped with my likely-bad-even-then handwriting...)

Today's sad thought: I wonder if there's any easy, request efficient (and bandwidth efficient) way to offer git repos for cloning that won't be excessively ransacked by LLM crawlers.

(... because if there is maybe the modern way to publish a git repo is 'here is the readme, here is how to clone, if you want anything more clone and look, you can thank LLM crawlers for this'. Maybe a (static) text page with a few of the recent commits.)

Current status: working how to get the equivalent of OpenBSD pflog with Linux nftables. Unsurprisingly there are more moving parts and they are less automatic and less well documented.

Blog post: Implementing a basic equivalent of OpenBSD's pflog in Linux nftables https://utcc.utoronto.ca/~cks/space/blog/linux/NftablesImplementingAPflog
tl;dr: nftables 'log group N' for some N, then 'tcpdump -i nflog:N ...'. Actually logging things to disk is more complicated and annoying.

@JdeBP Now I know that EBCDIC has lower case first (lower numbered bytes) than upper case, which is interesting because I bet a lot of early sorting was 'byte order'. If I was energetic I'd try to hunt up discussions of how ASCII was designed to see if lower or upper case first was explicitly discussed.

I'd heard various stories of the origin of 'posh', so I looked it up in the OED via the university and it's more interesting than I expected:
* 1914: modern 'posh' as slang from soldiers in WW I, initially about superior clothing.
* 1890: 'dandy', in slang dictionaries, with a 1912 attribution as 'well dressed'.
* 1876: northern regional slang for 'a soft mass' or 'fragments produced by an impact'.
* 1830: slang for (small amounts of) money.

One can make some amusing theories from this.

@ireneista We have some industrial hearing protection things at work for when we're in our machine rooms, and I've definitely noticed that the frequency response of the reduction is not flat.

(I'm more picky about noise level that my co-workers so I'm the most common user of them. The machines rooms aren't loud-loud, just noisy enough that I'd rather have the quiet(er).)

Hot take: if people say data is oil, agree with them and then point out what they're saying. Which is that it's a relatively low value toxic substance that's hard to store safely or move around, and if you do things wrong it explodes, catches on fire, or contaminates your entire company, or all three at once. Do you want to store a bunch of barrels of oil in your building? Not if you're sane.

In re: https://infosec.exchange/@SwiftOnSecurity/114916823693937019

Dear (FreeBSD) vim: when I am using 'cat' to 'edit' a file so that I can paste things from one xterm terminal window to another, you have failed. Spectacularly. You had approximately one job.

If you email me as a result of one of my techblog posts to ask me for any opinions on your commercial product, and then you email me a couple of weeks later when I haven't replied, your email messages immediately go to the bottom of the pile and the odds there will ever be a reply go almost to zero.

(This is not quite spam because I'm sure the person involved actually read my post and wrote the email with sincerity, but. I do not owe you any commentary at all & poking me is not how to get it.)

@oclsc The NFS server behavior seems to be long-standing (it's definitely in 22.04), and ls/Coreutils using listxattr() instead of other ways to probe for xattrs seems like a sensible switch (or maybe ls added support for noticing the presence of xattrs). The combination is the unfortunate thing, and if anything I blame the NFS server (which has been doing that since xattr support was added in 2020).

(The coreutils test matrix can only be so big, at some point you assume sane FSes.)

TIL that the version of ls in Ubuntu 24.04 will print errors if it can't get a list of a file's xattrs (used for, eg, 'ls -l'), and also that the Linux kernel NFS server will only let you list xattrs for objects you have read permissions on, although you can stat() them fine (of course) and this is not how local filesystems behave (and so 'ls -l' works fine on unreadable things on them).

In related news, we now have a patched local version of the 24.04 Coreutils package.