2008

Wed, Aug 20

<00:55 EEST> A recent trend with higher-end laptops seems to be a kind of Linux BIOS, into which you can boot near-instantly for simple web and office tasks, instead of waiting for a Redmond monstrosity to kick off. This PC World article interprets this nicely as a case of Linux taking over the world while the giant literally sleeps. It mentions a rather elaborate implementation with a separate ARM processor that can run Linux for days with the usual laptop battery.

I wonder if there are some half-sentient consumers asking themselves: what do I need Windows for, if Linux can do all this? Why doesn't everyone use Linux instead, when Windows is wasting all this power? Could I run this laptop with Linux only?

A related issue is the dominance of x86 over more sensible processor architectures, on which I read an x86 vs PPC roundup. It cleared up a lot of details of which I'd previously had a hunch. One important thread that emerged from the analysis is that the x86 market is like an ultimate race, where everybody wants to win no matter what, whereas PPC chips are based on solid engineering and science. As a result, PPC processors are somewhat more efficient per clock, significantly more efficient per watt, and easier to program for.

One of the key differences is the instruction length, variable in x86 and fixed in PPC. This means that x86 is often more space-efficient, and a code that takes up less memory can be faster if memory-bound. On the other hand, PPC instruction handling is much simpler, hence more efficient. An intriguing parallel is IPv6, which is usually marketed with its massively increased address space, but provides other, more technical benefits too. One of these is the fixed-length header that should make routers faster, despite the longer addresses involved.

Mon, Aug 11

<19:40 EEST> Read the book "Sudenluusta supertietokoneeseen", a history of computing by Jarmo Pulkkinen. Having read quite a lot of material on the topic, I'm naturally comparing this to a number of others. Unfortunately, it's somewhat apparent that the author lacks some key insights that would be expected from a mathematically minded scientist.

Despite some glaring problems, the overall structure and purpose works out pretty well. For example, I like the overarching theme of the duality between digital and analog computers. This often becomes apparent in that business and state people prefer the unforgiven precision of integers and digital methods. On the other hand, scientists acknowledge the inherent uncertainties in real-world quantities, and also prefer the flexibility and speed of analog computers such as sliderules.

An interesting issue, which the author omits, is the relation between analog computing and the floating-point capabilities of modern CPUs. While FP calculations take place in the digital realm, being deterministic and reproducible, they do have a limited precision. Users and programmers need to take this into account, and treat the FP unit as an analog computer in practice.

This comparison is even more interesting with instruction sets like SSE and Altivec; a lot of recent CPU development is directed towards faster and more parallel FP calculations. While it may mostly be the entertainment industry that's behind this choice, the end result is that you need to use parallel FP to get most of the CPU power. So we've ended up with a search of prime numbers that works best with SSE2. Though there's also the factor (pun intended) that this particular search is based on Fourier transforms :)

Another structural choice I like about the book is that it discards a strict chronology and a jumbled list of inventions with something in between. Each chapter is usually an independent chronology of one particular field of computing, though connections and big pictures are occasionally drawn. Coupled with the moderately artsy writing style of a historian, this makes the book a fluent and enjoyable read.

Fri, Aug 8

<01:04 EEST> Spent a couple of days modifying nunu for my own purposes. Basically removed everything that has to do with the CRT monitor, with surprisingly cool results. Thursday also marked the Pikkujattilainen's return to stage after a fortnight's break.