tHog

DIARY 2011

(2010)

DECEMBER

NOVEMBER

OCTOBER

SEPTEMBER

AUGUST

JULY

JUNE

MAY

APRIL

MARCH

FEBRUARY

JANUARY

2011

Wed, Mar 30

<19:16 EEST> I woke up this morning to my N900's rather annoying alarm tune, from the middle of a deep dream. I can still recall a colleague talking about methane, in relation to converting organic compounds into others. The world felt generally unreal for most of the morning, and this was not helped by the fact that one of my computers, in some ways one of the most central, was down.

After a few reboots, the culprit turned out my old friend, the 80-watt AC-DC adapter that has undergone capacitor surgery a couple of times before. The machine was soon back up and running, with only half of the usual number of GPUs and power adapters, but working nevertheless. A later dissection in the afternoon revealed yet another familiar sight, bulged and leaked output capacitors. The surprised ones were:

Some might say that my problems with this power brick have something to do with the concept of "bleeding edge". Bringing hardware to its limits, or as we Finns used to say during times of war: Rautaa rajalle! On the other hand, it is this one single PSU that takes the blame for a majority of my hardware woes, at least since buying it in 2004. I have even done a warranty return of another one of the same model, but fortunately for me, the store did not have the same models available, so they gave me a 110-watt model instead :) It is thanks to this incidende that I can actually power a HD5770 using my fanless rig, since the GPU draws over 100 W.

Considering the alternative of a new PSU, it is certainly worth replacing the capacitors a few more times. It is now less than two months since the previous breakdown, but that time a broken motherboard was likely involved; it even broke my Power Mac's PSU. I have not yet delved into its depths, but there is a possibility of merely having burned an internal fuse. Anyhow, it is again the time to whip up some solder and recall another call of war: Hakkaa päälle! In other words, time to hack on :)

Fri, Mar 18

<18:22 EEST> As Linux 2.6.38 was released on Monday, most of the news were highlighting the process group scheduler that was supposed to bring desktop interactivity to new levels. The way I understand it, a compile session that includes several CPU-heavy processes is counted as one group, whereas the X session is another, and thus it is not fighting for CPU time alone against dozens of compilers.

However, on my laptop things only got worse. Firefox was becoming quite a pain to use, while a niced process (dnetc) was hogging about half of the CPU time. Even if the new scheduler was not quite polished, the disregard for nice is something you just don't do.

In geek circles, there is a general contempt for software that tries to outsmart you, to decide how you should use your computer. The idea of grouping processes is a valid technical improvement in itself, but perhaps the automatic bit is doing something wrong. The traditional way of dealing with compilation and other CPU-intensive batch jobs in Unix is to use nice, and it has worked great so far. It is a simple manual way of setting priorities, and it is a little weird for the kernel to change these decisions for me.

Fortunately, it is merely a compile-time option and easily disabled. It may even work much better with more recent hardware; perhaps the overhead of extra scheduling work is to blame on slower machines. So far, I have not tried it on my stronger machines, as the ATI binary driver requires an older kernel.

Anyway, I can confirm that the sluggishness was indeed due to the process group scheduler, as 2.6.38 works like a charm without it. Things may even be a little faster compared to older kernels, but it may also be due to a new X server.

Mon, Mar 14

<21:09 EEST> Having just regained my interest in SIP with the N900, this GNU Free Call announcement was rather timely. From what I understand, the goal is a Free as in Freedom alternative to Skype, with centralized SIP servers replaced by the P2P SIP Witch. One open question I have is securing a unique identity, but there are generally cryptographic solutions to that.

For years, many a geek has wondered why pay extra for voice and SMS, since it is all simply data. There is of course the history of telephone networks which prioritize voice, and reliability is still a concern in mobile networks. Moreover, latency and jitter are notable inconveniences, particularly in 3G cell networks. This is about to change with 4G being a more unified technology, but it remains to be seen how "voice" data is handled relative to other data. Perhaps we will keep paying for QoS to get landline levels of reliability, and have the choice of cheaper "data calls" for less crucial situations. Hopefully with the unified identity of the same phone number, if so desired.

Naturally, it is hard to switch everything at once. Even this N900 does not have 4G connectivity. What is great about the different SIP solutions, and even Skype, is the bridging to traditional phone networks. People are starting to use these services for cheaper calls to "real" phones. Soon there are enough people using these that they can skip the phone network altogether.

Sat, Mar 12

<20:39 EEST> Finally putting my money where my mouth is, I got an N900 this week. It has been on my shopping list since before its release, but I never actually felt the need to own a smartphone. Not even now, but since Nokia is going down the drain with Microsoft, I felt it is time to buy the only proper Linux phone while they are still available. Incidentally, it is almost precisely 10 years since I bought my last phone, likewise a used Nokia.

Right now I lack the time and interest to fully explore its possibilities, but I do have some nice observations already. For example, having GSM voice, Skype and SIP integrated under the single calling interface. I do not have a mobile data contract at the moment, but at least while at home or any other WLAN, I can use my Saunalahti SIP account to call other phones more cheaply.

The keyboard is very nice for its size, though the three rows pose some serious limitations. Numbers and punctuation must be accessed via modifier keys, and the punctuations end up in non-standard places. Of course, at this size there is no touch typing anyway. The touch screen interface works well with fingers, though a stylus is also included.

Another piece of this week's "new" hardware is a Sapphire Radeon HD5870, intended for Bitcoin and possibly other number crunching applications. Unfortunately, it does not quite fit my kind of use. The windtunnel design seen in many graphics cards is probably great for ordinary cases, where the warm air must be pushed outside through the back panel. It is also robust and convenient to handle, compared to the twisty mazes of exposed heatpipes and heatsinks. However, it is not optimal for keeping cool and quiet in an open case, especially when used for days and weeks at a time. I have already turned my attention to other variants of the same GPU, though a small price premium is to be paid.


Risto A. Paju