WezM.net - All Articles

A Photo a Day in 2012

2012-01-21T06:18:00Z

This year I've decided to have a go at publishing a photo every day of the year. I'm not the first to do this, its also known by the name Project 365. I'm publishing my photos to the 2012 set on Flickr. The first week whilst I was still on Christmas holidays was pretty easy. Although I am missing a photo for Jan 2 as the photo I took got deleted whilst trying to restore my iPhone that got drowned in a river. Anyway after returning to work its been harder to find things to take pictures of but I haven't missed a day yet.

Some of my favourites from the first 3 weeks are:

Jan 8 — Eastern Spinebill

Jan 13 — CityLink

Jan 15 — Sulphur Crested Cockatoo

Jan 18 — Belvedere – Intense

Jan 19 — Macedon Primary School Plantation

Home NAS Powered by FreeBSD and ZFS

2012-01-21T06:08:00Z

Summary

Here's the tl;dr version: I sold my Mac Pro to fund building a home NAS. The result is a HP MicroServer with 4Gb RAM and 3 × 2Tb hard drives running FreeBSD from the system drive and a ZFS pool across the three 2Tb drives. Total cost: AU$731.67.

Rationale

Recently the Time Machine drive in my Mac Pro started to randomly disappear and Mac OS X would say that I had removed it improperly, which was not true given it was an internal drive still inside the machine.

I've seen this behaviour before and in that case it resulted in the drive being replaced due to its inability to complete a short S.M.A.R.T. scan. This drive (also a Samsung) was suffering a similar problem except that initiating the S.M.A.R.T. scan would actually cause it to disappear from the SATA bus. A check on the Samsung site showed that the drive was out of warranty so I was up for a replacement.

The Mac Pro wasn't getting used for much since I got a i7 powered Mac Book Pro. Its main duties involved storing my iTunes library, Aperture library and running my weather logger. It wasn't exactly a very energy efficient machine to run all the time. It would in fact keep the study warm overnight when the door was closed during winter.

There was also a problem with replacing the failing drive: I couldn't afford to do so. So I decided to move the weather logging to my ALIX board and sell the Mac Pro to fund building a home NAS. I was able to sell the Mac Pro very quickly on eBay for $1500 but gave myself a budget of $1000 for the NAS. I wanted the NAS to have reliable, redundant storage, which for me meant ZFS. This implied the new machine would need to run one of Solaris, illumos, FreeBSD, FreeNAS or SmartOS. The requirement to run one of these OS's ruled out an off the shelf NAS appliance.

I did a lot of research into different ways to build the machine and tried out all the OS options in virtual machines. I considered using basic PC hardware, MiniITX, HP MicroServer, etc. Each had its own pros and cons. The basic PC approach was possibly the cheapest but it was the largest. MiniITX was more expensive and choice of multi hard drive bay cases were limited. I ended up settling on the HP Proliant MicroServer running FreeBSD.

The end result. Click/tap to toggle inside view.

CD for size comparison.

Oblique view (excuse the finger prints).

The Build

The MicroServer is a neat little unit. It uses a low power dual core AMD Turion II CPU and comes with 2Gb ECC RAM and a 250Gb HD. I has 4 non-hot swappable hard drive bays all packaged up in a squat little box. I ordered mine with an extra 2Gb or RAM as ZFS likes to have plenty of RAM available to run well.

During my research hard drive prices sky rocketed due to floods in Thailand, however I was able to get some at pre-flood prices from ht.com.au. They have since put the price up ~$40 and placed order limits on them.

For the drives I chose 2Tb Seagate Barracuda Green's. They feature SATA 3 and a 64Mb cache and run at an atypical 5900RPM. These drives seemed to be a good balance across energy efficiency, noise, performance and price.

The final parts list ended up being rather diminutive:

1 × HP MicroServer (658553-371) + 2Gb extra RAM $336.82
3 × 2Tb Seagate Barracuda Green Hard Drives $394.85

The total cost ended up being $731.67, healthily under budget.

Installing the extra RAM.

Installing the hard drives.

Software

Installing FreeBSD and setting up the ZFS pool was very straightforward. I'm running the drives in a RAIDZ configuration, giving 3.6Tb of usable storage. I currently have two ZFS file systems on that. One in a normal configuration and the other for photos with copies=2 set.

The system all ran well for a few days however on the forth day one of the brand new drives failed and started making a terrible clicking, beeping noise. Fortunately HT replaced it very promptly and the replacement has been running fine since. During the time the failed drive was out for replacement the ZFS pool continued to run fine in its degraded state, with no data loss. Once the new drive was installed it was a simple matter of issuing zfs replace ada1 and it began the process of resilvering the data onto the new drive and all it has been running well since.

$ zpool status
  pool: storage
 state: ONLINE
 scan: resilvered 1.07T in 9h32m with 0 errors on Tue Nov 29 07:13:29 2011
config:

  NAME        STATE     READ WRITE CKSUM
  storage     ONLINE       0     0     0
    raidz1-0  ONLINE       0     0     0
      ada1    ONLINE       0     0     0
      ada2    ONLINE       0     0     0
      ada3    ONLINE       0     0     0

errors: No known data errors

After setting up the OS and file systems the only other thing I needed to so was make the storage available to other machines on the network. Since my house is all Macs I built netatalk via the FreeBSD ports collection to make the storage available via AFP.

With that done it's the sever shows up in the Finder via Bonjour and copying/accessing data is dead simple.

Little Linux Router Box

2012-01-21T04:33:00Z

After ongoing issues with maintaining a reliable Internet connection at home I decided to add a router box to the network in charge of assigning IP addresses and sharing our Internet connection with the rest of the network. I wanted something with at least two Ethernet ports so that all Internet traffic would flow through the device and allow bandwidth hogs to be identified at times when the connection appeared flooded.

After a bunch of research into appropriate hardware and software I decided on a PC Engines ALIX single board computer (alix2d13). The ALIX is a small board about the size of a CD case with the following features:

500Mhz AMD Geode (x86 compatible) CPU
256Mb RAM
2 × USB ports
3 × 10/100 Ethernet ports
Mini-PCI socket (for WiFi if desired)
RS-232 serial port
CompactFlash socket

Along with the board I ordered one of PC Engines cases (case1s2u) to go with it.

For the software I settled on OpenWRT. I chose it for a number of reasons:

Good support for ALIX boards.
Designed to run from Flash and read only file systems.
Great binary package manager that allows additional software to be installed.
Lightweight Lua based Web UI available (LuCI).

After learning to use the excellent OpenWRT build system I was able to build a custom CompactFlash image for the board to run. I also created a package for my weather logging software so that the ALIX can do all my weather station logging.

This setup has been extremely reliable. I've pushed all services on to it that my ADSL modem and AirPort base station used to be responsible for. Including maintaining the PPPoE connection to my ISP. I'd certainly recommend a set up like this to anyone who is looking for a small dedicated home router.

If you don't like the prospect of building your own custom OS image I'd also highly recommend m0n0wall, which is a FreeBSD derived router system. I ran this on the ALIX initially and it was very easy to get up an running (write image to CF card, boot from card - default settings worked fine but can be changed via a web UI) and very reliable.

Front with three LEDs. The behaviour of the LEDs can be customised in OpenWRT. I have the left one indication power and the middle one indicating activity on the WAN port. The other one is currently unused.

Top of case, showing size in comparison to a CD case.

The ALIX board itself with the following connected: WAN and LAN Ethernet, 8Gb MosKeyto USB flash drive, weather station USB cable and 32Mb CF card that the system runs off.

Fun With objc_setAssociatedObject and UIAlertView

2011-12-11T06:26:00Z

When it comes time to present errors or other messages in iOS with UIAlertView it is immediately obvious that a more convenient interface would involve the use of blocks. A search on GitHub shows just about every iOS developer has had the same thought and had a crack at it.

After reviewing the better options (I.e. those that actually had a README with more than a few lines of content) on GitHub it appeared that one of the challenges was how to handle memory. This stems from the wrapper becoming the delegate for the UIAlertView but not having a strong reference from the caller. I saw various solutions to this, most involving adding a extra retain call to keep everything around until the delegate methods were called.

Since my project is using ARC I looked for a solution that didn't involve marking the wrapper file as non-ARC and invoking retain. The solution I came up with was to use objc_setAssociatedObject. This Objective-C runtime function allows one object to be associated with another using various memory management strategies. I used this to associate the UIAlertView blocks based wrapper with the UIAlertView.

In the init method the wrapper instance is associated with its UIAlertView:

objc_setAssociatedObject(alertView, _cmd, self, OBJC_ASSOCIATION_RETAIN);

Note that I'm using the implicit second argument to the method, its selector, _cmd as the key for the associated object. This was suggested in a tweet by Bill Bumgarner.

Then in alertView:didDismissWithButtonIndex:, the association is removed, dealloc of the wrapper called as a result and the UIAlertView also released.

SEL key = @selector(initWithTitle:message:);
objc_setAssociatedObject(self.alertView, key, nil, OBJC_ASSOCIATION_RETAIN);

The full MIT licensed code is available as a gist on GitHub.

The iOS Launch Image is Not a Splash Screen

2011-12-07T20:29:00Z

At work we're currently having an iOS app that I'm working on styled by a designer. The second screen we were supplied was a pretty splash screen with a big logo on it. We responded saying that we didn't want to use the launch image like that. Our designer responded saying that was fine but the majority of the clients nowadays require heavy emphases on corporate branding therefore it was a standard practice to cater to those requirements. It's a shame that this is what people are are asking for since its is not the intended use of the launch image.

Justin Williams recently had the following clear-cut words to say on the topic of splash screens in his On Magazines and the iPad article (emphasis from the article):

Remember, kids. The first rule of mobile development is that no one gives a fuck about your brand. A splash screen with a giant logo is something that makes editors and marketing directors feel good, but to a user it just feels like a meaningless delay. You know that feeling of frustration you get each time there’s a 15-second preroll before a video on the web? That’s what a splash screen with logos and advertisements is.

The Apple iOS Human Interface Guidelines include the following suggestions:

Avoid taking space away from the content people care about. For example, displaying a second, persistent bar at the top of the screen that does nothing but display branding assets means that there’s less room for content. Consider other, less intrusive ways to display pervasive branding, such as subtly customizing the background of a screen.

Display a launch image that closely resembles the first screen of the application. This practice decreases the perceived launch time of your application.

Avoid displaying an About window or a splash screen. In general, try to avoid providing any type of startup experience that prevents people from using your application immediately.

Supply a launch image to improve user experience.

Avoid using your launch image as an opportunity to provide:

An "application entry experience," such as a splash screen

An About window

Branding elements, unless they are a static part of your application’s first screen

Because users are likely to switch among applications frequently, you should make every effort to cut launch time to a minimum, and you should design a launch image that downplays the experience rather than drawing attention to it.

The launch image is designed to make the perceived launch time of you app feel faster by showing something resembling the interface that will be loaded as quickly as possible. Displaying a logo does nothing but draw attention to how quickly your app loads and adds nothing to the user's experience.

Add a Custom Search Engine to Safari

2011-09-24T05:34:00Z

In my ongoing efforts to distance myself from Google I started using DuckDuckGo as my default search engine. I tried the DDG Safari extension but didn't really like it, especially since it needed to add an entire new toolbar to the browser. Other suggestions for adding DDG to Safari involved hacking the binary or other extensions, which I wasn't interested in.

Update: There an easier way to achive what I describe below: Just add a hosts file entry for search.yahoo.com that points at DuckDuckGo's IP address and set you search engine to Yahoo! in Safari. See DuckDuckGo for the full instructions.

My solution to the problem was to hijack the Bing option in the default search box for use with DDG. Note: This solution assumes you will never want to go to the www.bing.com domain, not a problem for me. The steps to implement it are:

Add an entry for www.bing.com to /etc/hosts that points the domain to your local machine:

127.0.0.1    www.bing.com

Next configure an Apache virtual machine to respond to the www.bing.com domain and redirect the request to Duck Duck Go (or your search engine of choice). This works because DDG accepts the search query in the same query string parameter, q, as Bing and ignores the other Bing related params.

/etc/apache2/extra/httpd-vhosts.conf:

<VirtualHost *:80>
    ServerName www.bing.com
    RewriteEngine On
    RewriteRule ^/search(.*)$ http://duckduckgo.com/$1 [redirect,last]
</VirtualHost>

For this to work you will need to have "Web Sharing" enabled in the Sharing preferences pane and have the the following line included in /etc/apache2/httpd.conf:

Include /private/etc/apache2/extra/httpd-vhosts.conf

Restart Apache (sudo apachectl graceful) and set your search engine to Bing in Safari. Do a search and you should end up at the DDG results. One of the neat features of DDG is its !bang syntax, which allows you to search to 100s of sites directly. One of which is !g for those times when you need to fall back on Google.

Strip Trailing Whitespace in Xcode 4

2011-08-27T09:36:00Z

Xcode has a bad habit of adding trailing whitespace to code, which is a pet peeve of mine. It introduces irrelevant changes into diffs and is particularly glaring when commiting code with GitX, which highlights it in red. In TextMate and now Vico I have bound the strip trailing whitespace in current document action to ⌃⌥⌘S (control-option-command-s). This makes it quick and easy. Xcode makes adding these types of actions a bit harder but I came up with a solution that uses the Behaviours functionality in Xcode 4. This is how I did it:

Create a shell script that invokes sed on each git tracked file that is modified. Since Xcode doesn't tell the script what the current file was and I didn't want to run sed over every file every time, processing tracked files with modifications was the best solution I could come up with. Note that the script also only prcocesses .m and .h files. I have my copy of the script in ~/Documents/strip.sh, be sure to give it the execute permission.

Next you need to add a behaviour to Xcode. Go to the Xcode preferences and click the Behaviours section, then click the + button. Name the behaviour and give it a keyboard shortcut. In the right pane check Run and choose the script you saved above. That's it. Now you can kill off that nasty whitespace with ease.

Monochrome Thumbnails

2011-08-27T08:54:00Z

Ever since this version of my site went live I've been meaning to post about the generation of the monochrome thumbnails on the front page. When I was building the site I wanted to include recent items from my Flickr photostream. However with the predominantly monochrome design I didn't like the look of all the colour they added. So I looked into having monochrome versions of the thumbnails shown.

The first thing I investigated was Javascript image processing libraries. The only main contender in that space was Pixastic. It worked but its cross browser support wasn't great at the time. So I moved on to writing a tool that I could run locally.

I decided the tool should generate a single image with both the colour and monochrome versions in it (a technique known as spriting). Using a single image means only one HTTP request instead of forty. Being a programmer with an interest in Cocoa programming I created a small command line tool that used the Flickr API to get the details of the last twenty images in my photostream, fetch them and then use Core Image to convert them to monochrome, add both the colour and monochrome versions to the output image and finally save the result.

This worked great, however when I stopped hosting my site on my Mac I thought I would need to run the process periodically on my Linux server, which ruled out Core Image. I took this as another opportunity to learn something new and rewrote it in Lua using imlib2 bindings. I had to make some additions to the imlib bindings, which are published on GitHub.

I used the Lua version for some time but never bothered to set it up on the server. I wasn't uploading photos all that frequently and it was simple enough to run locally and rsync the result. I've recently switched back to the Mac version as it was simpler to get up and running on my new laptop.

The code is on GitHub if anyone wants to do something similar. The Cocoa version is on the master branch, the Lua version is on the lua branch.

The current version of the processed thumbnails is shown below.

Touch Typing

2011-06-06T07:15:00Z

Being a professional programmer I spend plenty of time in front of a keyboard. Unfortunately I never learned to touch type. Instead I developed my own style of typing that uses all my fingers but requires me to look at the keyboard. I've made a couple of attempts over the years to break this habit but always reverted to my old ways, primarily due to the large speed difference.

The topic came up during our Monday meeting today and it renewed my desire to break the habit. I've decided that committing publicly may help motivate me.

As the first step in this process I've painstakingly typed this whole post without looking at my hands. It's taken a long time, let's hope that improves quickly.

Polyphasic Sleep Experiment Over

2011-03-13T10:34:00Z

For anyone following along you may have noticed a lack of updates on my polyphasic sleep schedule. Sadly, as is frequently the case it seems when people document their experiences with a sleep schedule, mine has also come to an end. There were a number of factors, but ultimately getting sick as a result of the schedule was the tipping point.

In the weekend of my sixth week I attended a friends house warming. I had intended to leave in time to be home for my core sleep however due to the nominated chef of the eventing arriving late that wasn't the case. I tried taking my evening nap in the car but I didn't sleep properly. When I got home I slept for the core three hours.

The following day I had a family lunch to attend. This time I had to take my midday nap in the car and I didn't sleep in that one either, however I did feel refreshed for a couple of hours. My evening nap coincided with the trip home (my Dad was driving). Again this one was of little value.

During Monday I came down with cold like symptoms as well as a persistent headache. I went to bed early Monday night but got up on time. However after only a couple of hours I was nodding off. I grabbed another nap but still felt super tired. I slept for another hour and half but didn't feel well enough to go to work. I tried doing some reading but was falling asleep so I went to bed again and slept from 10 until 1. After that sleep I felt heaps better and the cold symptoms had greatly reduced. Tuesday night I resumed a monophasic schedule and returned to work on Wednesday.

On Wednesday it was great to not feel tired and significantly, be able to remain awake and productive on the train. I also felt more productive at work since I felt more switched on and not just fighting to remain awake.

Going into the experiment I underestimated the social aspect. I didn't think I had that many social events these days and that those I did have I could work around. This proved more difficult than I thought.

If I were to do it again I would not be a strict around only having a three hour core sleep and three naps. I think that once I'd experienced the shift from light to deep naps I would take a nap whenever I felt it was needed (within reason). The problem with sticking rigidly to the schedule is that there's no leeway for bad naps or anything like that.

So all in all it was an interesting experiment but for now I don't think it works for me long term. I've lost about three hours of useful time but I am enjoying being able to go the whole day without napping, not worrying about where I'll be, come nap time and not having to worry about being invited out to things. Its also great to have a pleasant, productive train ride again.