mgaunard

13 ms·

301.
by mgaunard 10mo ago
the whole point of TCP is that it is a stream of bytes, not of messages. The problem is that this is not in practice quite what most applications need, but the Internet evolved towards UDP and TCP only. So you can have message-based if you
302.
by mgaunard 10mo ago
WebSocket is full of web-tech silliness; you'd be better off doing your own framing.
303.
by mgaunard 10mo ago
you can already send fragmented data in one system call without copying it to a single buffer.
304.
by mgaunard 10mo ago
I've always found Nagle's algorithm being a kernel-level default quite silly. It should be up to the application to decide when to send and when to buffer and defer.
305.
by mgaunard 10mo ago
It's starting with the assumption that the simulation would reproduce the universe perfectly; this eliminates a lot of possibilities. Many would expect that the parent universe would be more sophisticated, potentially with more dimensi
306.
by mgaunard 10mo ago
The author published the bandwidth of its algo, it's one fifth of a typical memory bandwidth (it's not possible to go faster than memory obviously for this benchmark, since we're assuming the data is not in cache).
307.
by mgaunard 10mo ago
Running more code per unit of data does not make the code hotter or reduce the register pressure, quite the opposite...
308.
by mgaunard 10mo ago
That doesn't make sense; the search is doing on-the-fly normalization as part of its algorithm, so it cannot be faster than normalization alone.
309.
by mgaunard 10mo ago
In practice you should always normalize your Unicode data, then all you need to do is memcmp + boundary check. Interestingly enough this library doesn't provide grapheme cluster tokenization and/or boundary checking which is one o
310.
by mgaunard 11mo ago
Converting to decimal is just as absurd. All you need is the final binary digit, which incidentally is the most optimal codegen, `v & 1`.
311.
by mgaunard 11mo ago
A much cooler approach would have been to generate the ASM from the same program, rather than generate a file from python and load that file from C++. The multi-stage build and filesystem are completely unnecessary. The technique actually h
312.
by mgaunard 11mo ago
You can just write the benchmark in such a way that the optimizations are not possible.
313.
by mgaunard 11mo ago
I feel the results say more about the testing methodology and inlining settings than anything else. Practically speaking all lambda options except for the one involving allocation (why would you even do that) are equivalent modulo inlining.
314.
by mgaunard 11mo ago
Even kubectl slows down to a crawl with a thousand deployments on the same cluster. The protocols are bad, as is the tech supporting them.
315.
by mgaunard 11mo ago
Those are all sub-par cloud technologies which perform very badly and do not scale at all. Some people would rather build their own solutions to do these things with fine-grain control and the ability to handle workloads more complex that a
316.
by mgaunard 11mo ago
Those are the main innovations tied to crypto trading. They do indeed have little to do with the blockchain or bitcoin itself, and do apply to any asset. There are actually several startups whose pitch is to bring back those innovations to
317.
by mgaunard 11mo ago
The only clear applications for AI in software engineering are for throwaway code, which interestingly enough isn't used in software engineering at all, or for when you're researching how to do something, for which it's not a
318.
by mgaunard 11mo ago
very little of the trading actually happens on the blockchain, it's only used to move assets between trading venues. The values of bitcoin are: - easy access to trading for everyone, without institutional or national barriers - high le
319.
by mgaunard 11mo ago
I'm on Linkedin a fair bit and I didn't see a single article about this.
320.
by mgaunard 11mo ago
You'd have the same problem if you bought your own hardware. You don't throw it away and replace it until it's been amortized.
321.
by mgaunard 11mo ago
That sounds like a good deal, what providers offer this?
322.
by mgaunard 11mo ago
They need it at the beginning to get started quickly, then they don't but don't bother moving out. Not too different from how many other lines of business get their clients in the door.
323.
by mgaunard 11mo ago
A high-density cabinet in a datacenter costs $4k at most, including power and bandwidth. That's nothing compared to an average AWS bill.
324.
by mgaunard 11mo ago
Elasticity is very expensive, in practice people only use it for one-off jobs, preferably using the cheaper unreliable "spot" instances (meaning the job must support being partially re-run to recover, which implies a complex job s
325.
by mgaunard 11mo ago
What AWS gives you is the ability to spin up dozens if not thousands of hosts in a single click. If you run your own hardware, getting stuff shipped to a datacenter and installed is 2 to 4 weeks (and potentially much longer based on how eff
326.
by mgaunard 11mo ago
Pricing in AWS is heavily dependent on whether you reserve the instance and for how long. In my experience, if you reserve a bare metal instance for 3 years (which is the biggest discount), it costs 2 times the price of buying it outright.
327.
by mgaunard 11mo ago
tl;dr: the implementation that is designed for fairness has lower standard deviation under contention, but otherwise performs slightly worse. Nothing too surprising.
328.
by mgaunard 11mo ago
Mutexes are a problematic pattern that doesn't compose, see the article.
329.
by mgaunard 11mo ago
it's a mutex iff it's acquiring a resource exclusively. which you don't need to do for synchronization of coroutines since you can control in which order things are scheduled and whether that's done concurrently or not.
330.
by mgaunard 11mo ago
An IP costs $50, or $0.50 per month if leasing.

More ›