HN user
sysctl vm.overcommit_memory=2. However, programs for *nix-based systems usually expect overcommit to be on, for example, to support fork(). This is a stark contrast with Windows NT model, where an allocation will fail if it doesn't fit in the remaining memory+swap.
I was talking specifically about double-precision floats. Single-precision floats can represent every integer up to +-2^24.
I have a way simpler explanation. IEEE 754 double can only represent integers up to 2^53 without precision loss, so if you naively average two numbers greater than 2^52, you get an erroneous result.
It just so happens that 2^52 nanoseconds is a little bit over 52 days.
I've seen the same thing with AMD CPUs where they hang after ~1042 days which is 2^53 10-nanosecond intervals.
Yes, you can chuck your SSD into a freezer. Data retention time increases exponentially in lower temperatures, so keeping it in a regular +4C fridge is enough to extend retention by decades.
Just remember to heat up the disk before writing and after storage.
https://www.ni.com/en/support/documentation/supplemental/18/...
Powershell is mostly case-insensitive, and most of the core cmdlets have short aliases. Try `Get-Alias` (or `gal`) to learn more.
You can use streaming WALs, like Litestream does. It is a pretty simple way to do incremental backups.
Yes, there is the technical appendix to the 2019 judgement.
https://www.jfsa.org.uk/uploads/5/4/3/1/54312921/technical_a...
It really doesn't help that it's written in ancient K&R C, but if you spend ten or so minutes just staring at it, familiar shapes and patterns start to appear. (Give it a try!)
Incidentally, it's in line with how APL code looks like an alien artifact at first, but you get used to it fast if you have spatial reasoning to wrap your head around reshaping and transposing.
In datacenters, you're mostly limited by the power (and thus cooling). Most commercial DCs only let you use up to about 10kW per rack. For standard 40U racks it's just 250W/RU, give or take.
There are niche expensive datacenters with higher power density, but as it stands, exotic multi-kW hardware at scale makes sense if you either save a ton on per-node licensing, or you need extreme bandwidth and/or low latency.
Zenfone 10 is only marketed as a compact phone. In reality, it is just slightly narrower than a Samsung S23 (146.3 x 70.9 x 7.6).
The shining example is https://grafana.wikimedia.org .
why do we measure time in hours, days, weeks, months, and years rather than in decaseconds, hectoseconds, kiloseconds, megaseconds, etc.?
We don't use them generally because at the time of French Revolution, due to taxes, standardizing the state units of measurement of physical goods was a much more pressing concern than time. (If I were to guess, there were no work hours limits, and thus it just hadn't crossed the popular mind). Weeks were changed to have ten days though.
That's not to say that they hadn't tried: decimal time was mandatory for a few years before they realized that there were too many clocks around, and pushing decimal time can actually turn people hostile against the then-new metric system.
We still kinda sorta use them in form of fractions of Julian days.
How long do you keep the transaction logs before rewriting them?
I too had a few EC2 instances go down with signs of being severed from the EBS in the recent couple of weeks; mine were in eu-west.
I'm pretty sure Lovecraft was at most a passing thought. The more immediate context would be the Putin's internet "AMA" in 2006[1], where "how do you feel about the awakening of Cthulhu" became a meme after becoming the second most popular question. Through the years, it and its associated memes like "fhtagn", "will devour brains", lost their connection with the origin and became a part of the background meme noise. (Compare it with how you don't think about /b/ any time your rickroll someone.)
When operating a complex system like Reddit, you know that something is going to happen during a significant event, but you generally don't know exactly what will break first.
What's worse is that the system evolved as a response to the previous similar event, and what broke now is probably not what had broken in the past.
Individually disposable, yes. But if you have a cluster of those, and you powered them on at the same time -- as it often happens -- you're in for an exciting ride when your servers start rebooting almost simultaneously, give or take a few minutes.
There are always bugs in silicon, just like there are bugs in software. They mostly show up under "a highly specific and detailed set of internal timing conditions". There are 40 documented erratas on EPYC 7002s alone; there are 35 in the 13gen Intel CPUs, including, curiously, RPL038, "Processor Exiting Package C6 or C8 May Hang". Mobile ARM chip manufacturers are notoriously bad at documenting their bugs, so who knows how many they have.
This one is interesting because its preconditions are so trivial, and it will affect many more people than usual.
Precisely. Compare your comment with these excerpts from the article:
Windows 95 is reasonably fast in performance, and is not compromised by the arguably frivolous animation and eye candy features in Windows 98 and newer versions.
I prefer to have as much native compatibility as possible with DOS applications such as older games I own.
Windows 95 was released 13 years prior to the article, roughly the same time span as between Windows 7 release and today. Windows 7 is still installed on a single-digit percentage of computers. It is interesting to see that the reasons outlined in the article could be used almost word-for-word by someone justifying their use of Windows 7.
SSE came along much later. At the time it was common-ish to just slowly send HTML tags. I think at some point there was a gorgeous hack where they dynamically rendered and sent GIF frames with the messages of the chatroom.
A connection is unwarranted if AWS gets an abuse letter about it.
Intel has been making GPUs since roughly 2002, 845 chipset series. (There were earlier ones, but I'm
Datacenters should have good, reliable and redundant supplies of power and connectivity. It is much easier to build a data center near or in a population center where fiber and electricity substations are plentiful, than to build your own long-haul fiber and high voltage lines into the middle of nowhere.
Of course you can do the latter if you have unlimited capital, but most companies choose the path of least resistance.
Don't know what the walkability argument is supposed to mean though: Amsterdam is famously walkable and cyclable, yet there are quite a few data centers right next to the center of the city; same goes for London.
You may want to print multiple pages per sheet.
Yeah, network scaling bugs are the most fun. The one I liked the most was when after expanding a pool of servers, they started to lose connectivity for a few minutes and then come back a minute or so later as if nothing happened.
Turns out we accidentally stretched one server VLAN too wide, to roughly 600 devices within one VLAN within one switch. The servers had more-or-less all-to-all traffic, and that was enough to generate so many ARP requests and replies that the switch's supervisor policer started dropping them at random, and after ten failed retries for one server the switch just gave up and dropped it from the ARP table.
Of course the control plane policer is global for the switch, so every device connected to the switch was susceptible, not just the ones in the overextended VLAN.
Pirate video releases generate heaps of traffic, and they won't be on the news. Earthquakes (mild ones, not the "cities collapse" kind) cause people to immediately go online and check on their friends. So do missile strikes, but those tend to generate extremely popular videos as well. During the holy month of Ramadan, you can see a rapid and deep traffic drop in Muslim countries right at their local sunset, when people have iftar, breaking their daily fast. A country in North Africa shuts down their internet access completely during their school exams. Popular web infrastructure sometimes reroutes your requests to distant data centers, leading to request latency exploding together with your queue lengths. In summer, the morning user activity peaks later than in autumn because schools are on their summer breaks.
I bet every seasoned SRE has a few to add.
They are used indirectly through distance. At the time, the meter was defined as one ten millionth of the distance between the north pole and equator through the Paris meridian. That means that the meter corresponds to 1/100000 of a grad of latitude -- which is better read as "a kilometre is 1/100 of a grad".
This is symmetrical to the nautical mile, which is one minute of arc.
No, relatime applies to how often atime is updated, while lazytime controls how often all three file timestamps (atime, mtime and ctime) are written out to disk. They are orthogonal: you can have strictatime+lazytime to have accurate atime tracking that generates no disk IO on reads. The downside is, of course, if your system crashes, the non-persisted atimes will be unreliable.
PSI tells you whether you have a problem or not; it's up to you to use the techniques described here to find the cause and the reason.
Great write-up about the troubleshooting process!
Regarding the exact case, there is a slightly deeper issue. XFS enqueues inode changes to the journal buffers twice: the mtime change is scheduled prior to the actual data being written, and the inode with the updated file size is placed in the journal buffers just after. If the drive is overloaded, the relatively tiny (just a few megs) journal buffers may overflow with mtime changes, and the file system becomes pathologically synchronous. However, since 4.1something, XFS supports the `lazytime` mounting option that delays the mtime updates until a more substantial change is written. Without it, the journal queue fills up at roughly the speed of your write() calls; with it, at the pace of the actual data hitting the disk, so even in highly congested conditions your application can write asynchronously -- that is, until dirty_ratio stops your system dead in its tracks.