BGP, NATO, Fragmentation, Spindown, Intelligence Law, Glonass

Absolutely horrible routing again. Two data centers, both very near Frankfurt. (Less 1 ms to DECIX) But when you're networking between those two data centers, routing is DC1 - Frankfurt - Paris - Frankfurt - DC2. Now we've got 10ms latency on Frankfurt to Frankfurt traffic. - Duh! - That's with IPv4, when using IPv6 it's even funnier. Now the latency is double tot 20 ms. - Sigh. - Also one routing is quite interesting, from Frankfurt to Helsinki via London, sounds quite strange.
Funny news. NATO is worried about Russian activity around submarine cables. Let's see who's talking Operation Ivy Bells Hardly anything new? Often when you hear news like this, it's something they've been doing for decades and now they're worried someone else might be doing the same. Which kind of sounds funny. No I don't mean that it wouldn't be a problem at least for somebody. But still the setting is just like from the movie. Somebody stole the car, we just stole.
Great list of very good Cloudflare blog posts to read. Especially this Qualcomm ARM vs Intel X86 was really interesting post. Another very interesting article is this one about global request rate limiting.
This is great article about drawbacks of fragmentation, even if you would have 0 latency 4K random read (giving same performance as sequential read with Queue Depth 1). It's even then bad thing, so SSD won't fix fragmentation problem.
Leaky Bucket Algorithm. Sounds really inefficient for rate limiting, unless some optimizations are done. First optimization I came up with immediately when thinking about the problem, is only implementing leak as function of time, which is triggered only when bucket is updated. This avoids all the unnecessary updates the the bucket data during during leak and reduced amount of wasted I/O by roughly 50%. But I simply use for rate limiting buckets which are usually created when first hit comes, and emptied at specific time intervals. Of course using time since last update to reduce leak after that would be trivial to add. Also if bucket has been completely filled during that time, it's not getting updated anymore at all, and requests are discarded as early as possible in the processing pipeline to reduce tied and required resources. Also after the bucket has been full for some time, it gets deleted. Because there's no reason to store information about empty bucket. Generally when implementing anything, I'll try to avoid reads, but especially writes as much as possible.
HDD Spindown issues TOSHIBA DT01ACA300 just refuses to spindown, with built in timer. of course hdparm -y works well. Sure, I've set APM level to 127, won't make any difference after all. I just sigh, hate people who go and write so crappy firmware that it seriously annoys everyone. - Thank you for that. - Then it's so typical that someone made horrible crap code, and then there's just more crap code and solutions work around that original crap, which causes even more issues and problems. - Oh well, yeah, life as usual.
About new upcoming Finnish Intelligence Law. There were mentions that intelligence subjects, wouldn't be informed that they've been targeted. Ok, that's totally obvious. But the biggest joke was, that you're allowed to ask, if you've been targeted. That's kind of funny. Why would anyone without anything to hide do such request? And if you do, isn't it obvious that it's good to check your activities in at least past year and keep you under light monitoring for year to come? That's quite pretty trap for anyone. Of course there are cases, when they're not informing you anyway, even if you ask for it. But the whole process sounds kind of trap, even if it's supposed to do things more transparent.
GLONASS will get 2.0 firmware update, which will significantly increase navigation accuracy.

2019-04-14