Posts by Richard Tobin 375 publicly visible posts • joined 13 Aug 2007
They used to sell a two-volume edition of OED, photographically reduced, which came with a powerful magnifying glass. These days you may be able to access it free online through your public library.
Edit: I see they sell a single-volume edition now.
"If the *same kernel* does such a bad job given nearly 100 Arm cores but does fine with just half a dozen x86 cores, it suggests that the problem is not just in the OS."
Not at all. it suggests that the OS has a scheduling algorithm that works for a dozen x86 cores, and doesn't work for 100 Arm cores.
"Click through, read *all* of his blog posts about this, and maybe ask him."
Well, no. If I had time to do that, why would I bother reading the Register? But there are plenty of clues in the parts quoted in the article. For example, "But at the same time, 100 percent load on all cores means you cannot listen to music on Spotify or watch online videos, etc.": Why is the scheduler putting 100% load on all the cores when it doesn't have the resources to handle the interactive tasks? Just because there are 100 cores doesn't mean you have be using them all, all the time.
-- Richard
"But at the same time, 100 percent load on all cores means you cannot listen to music on Spotify or watch online videos, etc. All that because the CPU cores are occupied by the build processes."
This is nothing to do with the processor. It's bad scheduling by the operating system.
"The fake CAPTCHA prompts the user to open Spotlight on their Mac, and then paste a "verification code" into the search feature. The phony code is a curl command, and as soon as the victim hits Enter and executes it on their computer [...]"
But typing a command into Spotlight search doesn't execute it. And later the article refers to pasting a command into Terminal. There's something missing here.
"good C programmers know to fear the aggressive optimization of Boolean logic, which can be hazardous to their finished products". The optimizations described in the linked article are quite different from the problems facing cryptographers. The optimizations in the linked article are just wrong, and a compiler that does them is not conforming to the C standard. The problem for cryptographers is that they want to specify behaviour that isn't part of the C language, namely timing.
The "AI Winter" was not circa 1984. That was the time of the Japanese Fifth Generation project, the UK Alvey program, the heyday of Lisp machines, Prolog, and expert systems. The AI winter was about a decade earlier, and in Britain was largely associated with the Lighthill Report of 1973.
In the 1970s we had a PDP 11/40 that was turned on every morning, and it too would often not start in cold weather. We were told that this was because the 14" platters of the RK05F disk drives would contract until they were no longer within the tolerance for track positioning.
(The RK05F was a fixed version of the removable RK05 and was able to have twice the track density because it removed the manufacturing variability between disks.)
Bear in mind that in those days people were much probably much less inclined to treat numbers as decimals in general. We would write £1 2s 6d, 5' 8", 2lb 4oz.
During the run-up to decimalization in 1971 one of the things that they (the government, teachers, the BBC...) tried to drum into us was that a pound and five pence should be written £1.05 and not £1.5.
We had a DECwriter as the console on a VAX 11/750 running Berkeley unix. Every minute a cron job printed the date so that we could tell when any output had happened.
One day we suddenly started getting an error for every command - something like "out of processes". We found that the console had run out of paper, and several hundred "date" processes were stuck waiting for their output to complete.
A fine example of this is that many Canon printers can't be connected to your wireless network if the network implements 802.11r, a protocol to speed up roaming between access points. The symptom is particularly bizarre - the printer just doesn't prompt you for the wifi password. The workaround is to turn off 802.11r, configure the printer, and turn it on again.
