https://github.com/Overv/vramfs

Oh, it’s a user space (FUSE) driver. I was rather hoping it was an out-of-tree Linux kernel driver, since using FUSE will: 1) always pass back to userspace, which costs performance, and 2) destroys any possibility of DMA-enabled memory operations (DPDK is a possible exception). I suppose if the only objective was to store files in VRAM, this does technically meet that, but it’s leaving quite a lot on the table, IMO.

If this were a kernel module, the filesystem performance would presumably improve, limited by how the VRAM is exposed by OpenCL (ie very fast if it’s just all mapped into PCIe). And if it was basically offering VRAM as PCIe memory, then this potentially means the VRAM can be used for certain RAM niche cases, like hugepages: some applications need large quantities of memory, plus a guarantee that it won’t be evicted from RAM, and whose physical addresses can be resolved from userspace (eg DPDK, high-performance compute). If such a driver could offer special hugepages which are backed by VRAM, then those application could benefit.

And at that point, on systems where the PCIe address space is unified with the system address space (eg x86), then it’s entirely plausible to use VRAM as if it were hot-insertable memory, because both RAM and VRAM would occupy known regions within the system memory address space, and the existing MMU would control which processes can access what parts of PCIe-mapped-VRAM.

Is it worth re-engineering the Linux kernel memory subsystem to support RAM over PCIe? Uh, who knows. Though I’ve always like the thought of DDR on PCIe cards. All technologies are doomed to reinvent PCIe, I think, said someone from Level1Techs.

Ok, I have to know: how is this done, and what do people use it for?

For my own networks, I’ve been using IPv6 subnets for years now, and have NAT64 translation for when they need to access Legacy IP (aka IPv4) resources on the public Internet.

Between your two options, I’m more inclined to recommend the second solution, because although it requires renumbering existing containers to the new subnet, you would still have one subnet for all your containers, but it’s bigger now. Whereas the first solution would either: A) preclude containers on the first bridge from directly talking to containers on the second bridge, or B) you would have to enable some sort of awful NAT44 translation to make the two work together.

So if IPv6 and its massive, essentially-unlimited ULA subnets are not an option, then I’d still go with the second solution, which is a bigger-but-still-singular subnet.

I’ve personally witnessed the seventh entry to this meme: lines of code over 1000 columns wide.

I try to forget, but the horror never fades.

Does anyone want to even guess what time this clock is supposed to be indicating?

For a link of 5.5 km and with clear LoS, I would reach for 802.11 WiFi, since the range of 802.11ah HaLow wouldn’t necessarily be needed. For reference, many WISPs use Ubiquiti 5 GHz point-to-point APs for their backhaul links for much further distances.

The question would be what your RF link conditions look like, whether 5 GHz is clear in your environment, and what sort of worst-case bandwidth you can accept. With a clear Fresnel zone, you could probably be pushing something like 50 Mbps symmetrical, if properly aimed and configured.

Ubiquiti’s website has a neat tool for roughly calculating terrain and RF losses.

I don’t have an answer for your woes, but MTU issues are notoriously difficult to investigate and mitigate, as Cloudflare found out: https://blog.cloudflare.com/increasing-ipv6-mtu/

If you’re using SLAAC for auto IP assignment, then the resulting EUI-64-based address would be essentially static, based on the premise that your MAC address and local subnet prefix don’t change. Privacy extensions night get in the way, as well as Android’s randomized MAC feature, but those are adjustable.

Concrete example of threat modeling: if someone found out I was using Signal, for any reason at all, would that cause problems for me?

If yes, then Signal is not a good option. If no, then Signal may be appropriate. Why? Because in their documentation, they explicitly state that while messages are confidential, the fact that you’re using Signal cannot be hidden, and so they don’t make that guarantee.

Tbf, can’t the other party mess it up with signal too?

Yes, but this is where threat modeling comes into play. Grossly simplified, developing a threat model means to assess what sort of attackers you reasonably expect to make an attempt on you. For some people, their greatest concern is their conservative parents finding out that they’re on birth control. For others, they might be a journalist trying to maintain confidentiality of an informant from a rogue sheriff’s department in rural America. Yet others face the risk of a nation-state’s intelligence service trying to find their location while in exile.

For each of these users, they have different potential attackers. And Signal is well suited for the first two, and only alright against the third. After all, if the CIA or Mossad is following someone around IRL, there are other ways to crack their communications.

What Signal specifically offers is confidentiality in transit, meaning that all ISPs, WiFi networks, CDNs, VPNs, script skiddies with Wireshark, and network admins in the path of a Signal convo cannot see the contents of those messages.

Can the messages be captured at the endpoints? Yes! Someone could be standing right behind you, taking photos of your screen. Can the size or metadata of each message reveal the type of message (eg text, photo, video)? Yes, but that’s akin to feeling the shape of an envelope. Only through additional context can the contents be known (eg a parcel in the shape of a guitar case).

Signal also benefits from the network effect, because someone trying to get away from an abusive SO has plausible deniability if they download Signal on their phone (“all my friends are on Signal” or “the doctor said it’s more secure than email”). Or a whistleblower can send a message to a journalist that included their Signal username in a printed newspaper. The best place to hide a tree is in a forest. We protect us.

My main issue for signal is (mostly iPhone users) download it “just for protests” (ffs) and then delete it, but don’t relinquish their acct, so when I text them using signal it dies in limbo as they either deleted the app or never check it and don’t allow notifs

Alas, this is an issue with all messaging apps, if people delete the app without closing their account. I’m not sure if there’s anything Signal can do about this, but the base guarantees still hold: either the message is securely delivered to their app, or it never gets seen. But the confidentiality should always be maintained.

I’m glossing over a lot of cryptographic guarantees, but for one-to-one or small-group private messaging, Signal is the best mainstream app at the moment. For secure group messaging, like organizing hundreds of people for a protest, that is still up for grabs, because even if an app was 100% secure, any one of those persons can leak the message to an attacker. More participants means more potential for leaks.

When I see E2EE and XMPP mentioned, I think of this blog post by Soatok, outlining some very odd cryptographic choices in XMPP + OMEMO: https://soatok.blog/2024/08/04/against-xmppomemo/

I would very much like to see a richer playing field than just Signal for private messaging, but it’s a tough nut to crack. For exactly which aspect that turns me away from XMPP for E2EE, I think this nails it down:

you only need check whether OMEMO is on by default (it isn’t), or whether OMEMO can be turned off even if your client supports it (it can).

When the competition is Signal, these sorts of details matter a lot.

If only one side of the switch/points remain, depending on the type of crossing and condition of the wheels, there’s a chance that the trolley’s right side wheels can jump over the switch and continue straight forward, even as the switch is set to diverge onto the non-existent siding.

Or it could derail but continue barreling forward anyway. But trolleys don’t tend to be going that fast.

Let me make sure I understand everything correctly. You have an OpenWRT router which terminates a Wireguard tunnel, which your phone will connect to from somewhere on the Internet. When the Wireguard tunnel lands within the router in the new subnet 192.168.2 0/24, you have iptable rules that will:

• Reject all packets on the INPUT chain (from subnet to OpenWRT)
• Reject all packets on the OUTPUT chain (from OpenWRT to subnet)
• Route packets from phone to service on TCP port 8080, on the FORWARD chain
• Allow established connections, on the FORWARD chain
• Reject all other packets on the FORWARD chain

So far, this seems alright. But where does the service run? Is it on your LAN subnet or the isolated 192.168.2.0/24 subnet? The diagram you included suggests that the service runs on an existing machine on your LAN, so that would imply that the router must also do address translation from the isolated subnet to your LAN subnet.

That’s doable, but ideally the service would be homed onto the isolated subnet. But perhaps I misunderstood part of the configuration.

I did indeed have a chuckle, but also, this shouldn’t be too foreign compared to other, more-popular languages. The construction of func param1 param2 can be found in POSIX shell, with Bash scripts regularly using that construction to pass arguments around. And although wrapping that call with parenthesis would create a subshell, it should still work and thus you could have a Lisp-like invocation in your sh script. Although if you want one of those parameters to be evaluated, then you’re forced to use the $() construction, which adds the dollar symbol.

As for Lisp code that often looks like symbol soup, like (= 0 retcode), the equal-sign is just the name for the numerical equality function, which takes two numbers. The idea of using “=” as the function name should not be abnormal for Java or C++ programmers, because operator overload allows doing exactly that.

So although it does look kinda wonky for anyone that hasn’t seen Lisp in school, sufficient exposure to popular codebases and languages should impart an intuition as to how Lisp code is written. And one doesn’t even need to use an RPN calculator, although that also aids understanding of Lisp.

Addendum: perhaps in a century, contemporary programmers will find it bizarre that C used the equal-sign to mean assignment rather than equality, when the <= arrow would more accurately describe assignment, while also avoiding the common error of mixing up = and == in an if-conditional. What looks normal today will not necessarily be so obvious in hindsight.

This doesn’t answer OP’s question, but is more of a PSA for anyone that seeks to self-host the backend of an E2EE messaging app: only proceed if you’re willing and able to upkeep your end of the bargain to your users. In the case of Signal, the server cannot decrypt messages when they’re relayed. But this doesn’t mean we can totally ignore where the server is physically located, nor how users connect to it.

As Soatok rightly wrote, the legal jurisdiction of the Signal servers is almost entirely irrelevant when the security model is premised on cryptographic keys that only the end devices have. But also:

They [attackers] can surely learn metadata (message length, if padding isn’t used; time of transmission; sender/recipients). Metadata resistance isn’t a goal of any of the mainstream private messaging solutions, and generally builds atop the Tor network. This is why a threat model is important to the previous section.

So if you’re going to be self-hosting from a country where superinjunctions exist or the right against unreasonable searches is being eroded, consider that well before an agent with a wiretap warrant demands that you attach a logger for “suspicious” IP addresses.

If you do host your Signal server and it’s only accessible through Tor, this is certainly an improvement. But still, you must adequately inform your users about what they’re getting into, because even Tor is not fully resistant to deanonymization, and then by the very nature of using a non-standard Signal server, your users would be under immediate suspicion and subject to IRL side-channel attacks.

I don’t disagree with the idea of wanting to self-host something which is presently centralized. But also recognize that the network effect with Signal is the same as with Tor: more people using it for mundane, everyday purposes provides “herd immunity” to the most vulnerable users. Best place to hide a tree is in a forest, after all.

If you do proceed, don’t oversell what you cannot provide, and make sure your users are fully abreast of this arrangement and they fully consent. This is not targeted at OP, but anyone that hasn’t considered the things above needs to pause before proceeding.

I mean, at the USA average price of electricity of $0.13 per kWh, then for a halving of 70 Watts, it’s about 11 cents per day, or $40 per year. But at the California average price of $0.35, then the savings is 29 cents per day, or $107 per year.

That’s not small money, especially if it’s free to make these gains by ripping out unneeded functionality. But the point is taken that it’ll be hard to find savings from older hardware, which simply didn’t prioritize energy efficiency.

A Nintendo Wii would also work, as exemplified by this blog running on a NetBSD Wii.

But in all seriousness, the original comment has a point: using a mobile phone as a server is possible but also wastes a lot of the included hardware, like the cellular baseband, the touchscreen, and the voice and Bluetooth capabilities. Selling the phones and using the proceeds to purchase a used NUC or an SFF PC would give you more avenues to expand, in addition to just being plain easier to set up, since it would have USB ports, to name a few luxuries.

From my limited experience with PoE switches, how much power being drawn in relation to how much the switch can supply has a notable impact on efficiency. Specifically, when only one or two ports on a 48-port switch are delivering PoE, the increased AC power drawn from the wall is disproportionately high. Hence, any setup where you’re using more of the PoE switch’s potential power tends to increase overall efficiency.

My guess is that it has to do with efficiency curves that are only reasonable when heavily loaded for enterprise customers. In any case, if either of those two candidate switches meet your needs today and with some breathing room, both should be fine. I would tend to lean towards Netgear before TP-Link though, out of personal preference.

Link to the blog post with the background on why this was made : https://ericwbailey.website/published/you-must-listen-to-rfc-2119/

Insofar as the skills hierarchy that software engineers develop well after learning to write in a programming language, I’m left wondering what scenarios or industries are the most “vibe coding” proof. That is to say, situations that absolutely require from day 1 a strong sense of design theory, creativity, and intimate knowledge of the available resources.

Musing out loud, history has given us examples of major feats of software engineering, from the Voyager spacecrafts, to retro console games squeezing every byte of ROM for value, to the successful virtualization of the x86 instruction set. In these scenarios, those charges with the task has to contend with outerworldly QA requirements and the reality that there would be no redo. Or with financial constraints where adding an extra PROM would cascade into requiring a wider memory bus, thus an upgraded CPU, and all sorts of other changes that would doom the console before its first sale. Or having to deal with the amazing-yet-arcane structure of Intel’s microchip development from the 80s and 90s.

It is under these extreme pressures that true diamonds of engineering emerge, conquering what must have appeared to be unimaginably complex, insurmountable obstacles. I think it’s fair to say that the likes of NASA, Sony and Nintendo, and VMWare could not possibly have gotten any traction with their endeavors had they used so-called “vibe coding”.

And looking forward, I can’t see how “vibe coding” could ever yield such “ugly”-yet-functional hacks like the fast inverse square root. A product of its time, that algorithm had its niche on systems that didn’t have hardware support for inverse square roots, and it is as effective as it is surprising. Nowadays, it’s easy to fuzz a space for approximations of any given mathematical function, but if LLMs were somehow available in the 90s, I still can’t see how “vibe coding” could produce such a crude, ugly, inspirating, and breathtaking algorithm. In the right light, though, those traits might make it elegant.

Perhaps my greatest concern is that so-called “vibe coding” presents the greatest departure from the enduring ethos of computer science, a young field not too tainted by airs of station. This field, I like to think, does not close its doors based on socioeconomic class, on the place of one’s birth, or upon the connections of one’s family. Rather, the field is so wide that all who endeavor for this space find room to grow into it. There is a rich history of folks from all sorts of prior occupations joining into the ranks of computer science and finding success. The field itself elevates them based on what they contribute and how they solve puzzles.

What strikes against this ideal is how so-called “vibe coding” elevates mediocrity, a simulacra of engineering that produces a result without the personal contribution or logic solving to back it up. It is akin to producing artwork that is divorced from the artist’s experience. It embodies nothing.

To be clear, the problem isn’t that taking shortcuts is bad. Quite the opposite, shortcuts can allow for going farther with the same initial effort. But the central premise of “vibe coding” is to give off the appearance of major engineering but with virtually no effort. It is, at its core, deceitful and dilutes from bona fide engineering effort and talent.

Circling back to the earlier question, in my personal opinion, something like the Linux kernel might fit the bill. It’s something that is now so colossally large, is contributed to by an enormous user and developer base, and fills such a sizable role in the industry, that it’s hard to see how “vibe coding” can meaningful compete in that space.