aseipp

48 ms·

511.
by aseipp 3y ago
The intended purpose of both Server Push and Early Hints is basically "Lower overall latency by letting the client know early on that there is more data it might need in subsequent requests." But in practice Early Hints are much b
512.
by aseipp 3y ago
High-volume single-writer + WAL mode is perfectly doable for a lot of applications, but you have to keep the concurrency model in mind. I think of it more like a really advanced data structure library than a "database" in that sen
513.
by aseipp 3y ago
People retweet your post in order to draw the attention of their followers to it, which is often called a "dogpile" (or whatever). But blocking someone prevents them from retweeting you, so it's a decent (but imperfect) mitig
514.
by aseipp 3y ago
They have definitely in the past added features that they later released as part of product launches and then upstreamed, e.g. ARM64 backend or more recently -fbounds-checks (which I think will be in the new Xcode?) Other things I don'
515.
by aseipp 3y ago
Fraud vector. It's very easy to turn compute into dollars if the compute is free or stolen. Cloud compute is very cost ineffective for something like coin mining, so any actual legitimate miners on any chain are going to just run every
516.
by aseipp 3y ago
Continuous performance monitoring of a service, from its inception. I'm building a storage service using SeaweedFS and also a web UI for another project. One thing I'm looking at doing is using k6[1] in order to do performance str
517.
by aseipp 3y ago
The compute accelerator story on mainstream, non-patched Linux, with upstream software isn't that good at the moment. You're going to be waiting a while before you can do fun stuff like organize layers across the Neural Engine and
518.
by aseipp 3y ago
Well, Nix mostly solves that issue so well due to a continuous massive offering of blood and sweat though, let's not kid ourselves. It's a significant amount of work to package arbitrary software because arbitrary software does we
519.
by aseipp 3y ago
The DGX described here is a distributed system in the sense that many nodes, each with their own GPUs, exist and are part of the overall whole. They are connected over Infiniband and use RDMA in order to read/write memory across the cl
520.
by aseipp 3y ago
> To me, this why AVX didn't get widespread use AVX has seen very, very widespread use across mobile, desktop, gaming, and server platforms. It's over 10 years old. Probably every hand-optimized vectorized x86 routine in the pa
521.
by aseipp 3y ago
Well, context and history is also probably relevant because for a very long time there was a lot of (mostly unwarranted, at least IMO) fear surrounding Mono and or not it was something Microsoft would care about in the future (or other rand
522.
by aseipp 3y ago
I've been wondering how to integrate modern .NET Core into a custom build system (buck2) and was wondering similar things. There's this project I think is cool called bflat[1] that basically makes the C# compiler more like the Go
523.
by aseipp 3y ago
What you're asking for now exists, Zen 4 mobile SKUs allegedly ship a Xilinx design on the die for "AI Acceleration" (some of their Versal fabric over some weird bus), that has absolutely 0 external software consumers beyond
524.
by aseipp 3y ago
> but the WebAuthn spec (which is what passkeys are based on) doesn't require any contact info to be provided. This isn't an issue with the spec, it's an issue with account creation, account information, and recovery flow
525.
by aseipp 3y ago
Most source code packages, including their metadata, are generally small even as an aggregated compressed file, and it's very common to download a lot of them all at one time once you've resolved the necessary dependencies. It dep
526.
by aseipp 3y ago
> But strategically, AMDs approach to GPU-CPU memory fusion is probably going to be the technically stronger approach. Assuming it works. I mean if anything, Nvidia is already there and crushing it too. CUDA has a unified memory model on
527.
by aseipp 3y ago
Only if you're purely 100% compute bound by a wide margin versus the size of your working set. But in that scenario, you can just widen the memory interface, lower the clocks speeds, and you'll normally still come out ahead in eff
528.
by aseipp 3y ago
Really exciting to see this and the recent renewed interest in more expressive ACL systems re: policy and (alternatively) relational access control. The pedigree of Cedar is also really interesting to me, coming from the angle that Torlak w
529.
by aseipp 3y ago
The lack of DWCAS as a primitive in general is really annoying, C++ aside. RISC-V's -A extension has no form of it, either; you only get XLEN-sized AMO + LL/SC (no 2*XLEN LL/SC either!) It's one of those features that wh
530.
by aseipp 3y ago
The recent work on relational, datalog-inspired egraphs in PLDI this year ("Unifying Datalog and Equality Saturation") is actually interesting because it can solve cases like the y/x*x -> y identity example, by the power o
531.
by aseipp 3y ago
Frankly in the particular case where you have cascading downloads of small files, HTTP/1 is so unbelievably bad at that compared to HTTP/2 (especially in cases where the user-agent throttles TCP connections, like the limits of ~6
532.
by aseipp 3y ago
The author is plenty aware of them, I can assure you (I used to work with him on the Glasgow Haskell Compiler; most Haskell programmers are plenty familiar with the ML module system, it's a highly coveted feature by many.) Anyway, what
533.
by aseipp 3y ago
ZGC relies on the space of 64-bit pointer (and the size of the 64-bit virtual address space) for its pointer coloring algorithm, and so it is by definition incompatible with UseCompressedOops even at small sizes. I suppose it's possibl
534.
by aseipp 3y ago
It's basically impossible to answer that hypothetical right now, because it depends entirely on the choice of client software. And that's still something that's evolving; it's just that Apple/Google/MS have the
535.
by aseipp 3y ago
You can use it with Chromium on Linux, but there's two parts to WebAuthn: the user-agent, and the client device/software that holds the key. So, WebAuthn shows you a QR code which you can scan. That happens on the user agent when
536.
by aseipp 3y ago
I mean, you could have a bunch of other things in the GPU instead of those cores, but the reality is that ray tracing serves multiple target markets for GPUs very well, and actually meets a demand people have: better graphical quality. In c
537.
by aseipp 4y ago
Close, but the link is on Page 13, and it points here: https://github.com/mwillsey/egg-smol Unfortunately the naming is all a bit confusing, isn't it....
538.
by aseipp 4y ago
Just using `cargo build` will work as you've seen, but building a Nix package won't work right now. If you're still keeping up with this thread, though, I have code in this repository to build a copy of buck2 with Nix -- it&#
539.
by aseipp 4y ago
You should look at Buck2. One of the coauthors of that paper (Neil) is one of the lead developers, and unlike Buck1 the buck2 binary exposes a very tiny API that consumers then build off of, and define all their rules, using Starlark for ev
540.
by aseipp 4y ago
There are always rough edges on other platforms, but these machines are good workhorses -- and would be an excellent choice for further development, to smooth out aarch64 support. I wish these machines (or Apple Silicon with Asahi Linux) ha

More ›