Pet Pixels - Techie

SunSpider benchmarks updated

marcus@bointon.com (Marcus Bointon) — Wed, 19 Mar 2008 11:26:03 +0000

I updated my previous post on Sunspider benchmarks to include Safari 3.1, FireFox 3b4 and IE 8b1.

Microsoft finally gets it

marcus@bointon.com (Marcus Bointon) — Fri, 14 Mar 2008 16:19:49 +0000

I somehow missed Microsoft's announcement that (in a complete U-turn from previous announcements) IE8 will support web standards mode by default, and thus any broken sites will have to enable IE7 mode by a meta tag. So finally, IE will cease to be the albatross around the neck of the internet, and developers the world over will at last be able to write standards-compliant sites that work in all major browsers.

I had real trouble believing that MS had convinced so many prominent web standards advocates (here and here) that the previous option was in some way a good thing, when it essentially meant that MS expected 99% of the web to change in order to support the 1% (almost entirely intranets and thus of no public interest) that are so badly written that they couldn't survive a browser update.

I'm very happy to see this change of heart, which was a really unexpected thing to see from MS. They don't normally give a stuff about such things, so they fully deserve the adulation that their announcement is getting in the comments. It also vindicates the slagging I gave the authors of those articles promoting the evil meta tag!

So, Thank you Microsoft! I look forward to not having to do anything special for IE - you probably just doubled the world's web development productivity rate! Who knows - one day IE might be as good as Firefox or Safari...

Email Luddism

marcus@bointon.com (Marcus Bointon) — Sun, 02 Mar 2008 21:26:00 +0000

Here's a little rant I've been meaning to get out for a while.

Whenever the subject of email client support of some particular email feature comes up, someone always posts a comment to the tune of "HTML doesn't belong in email anyway, all email should be plain text". What they're saying is that they outright reject two important features: MIME and progressive enhancement. Given that MIME is what makes attachments (or any arbitrary binary data, attached or not) possible, I guess they can live without them too. It's the same technology that allows web servers to identify content types, so while we're painting with his particularly tarry brush, I guess we should remove CSS, javascript and images from HTML pages as well. That should keep them happy. With all those removed, we can all retreat to the comfort of the command line where our needs will be served admirably by the likes of the wonderful (no joke) elinks.

The whole point of the multipart/alternative data type is progressive enhancement. A client is free to select from the alternatives presented and render as best it can, with an option for manual selection (that is, as long as you don't use Outlook which doesn't believe in such things). This applies to the common text/plain > text/html combo as much as it would to text/plain > image/jpeg, or perhaps application/pdf > application/vnd.sun.xml.writer. Now if they restricted their comments to text/html only, I might have some sympathy, as that's just shoddy behaviour on the part of the sender. However, they usually prefer to throw out the baby with the bath water.

To conclude: MIME is a wonderful thing; some people use it badly; get over it.

iPhone icon test generator

marcus@bointon.com (Marcus Bointon) — Fri, 15 Feb 2008 16:15:31 +0000

I just came across this neat trick for providing custom icons (think favicon.ico, but with a reasonable size, better colour and a proper file format) for web pages for iPhone/iPod touch users. There seemed to be some debate over what exactly the native size is, so I built a test page to test it. The full-size icon image is also displayed on the page, but that's only there to show what the phone is starting with.
After twiddling with this test for a while, I came to the conclusion that there isn't a native size - it's somewhere between 59x59 and 60x60 - though 60x60 is about as close as you can get. This lack of native size is interesting, as it implies that the iPhone UI is using resolution independent rendering, which we know OS X can do.
Bigger sizes do scale more smoothly, but they're a waste of bandwidth and mean that you lose control of the exact appearance - photographic icons will look very nice, but anything involving single pixels lines will probably suffer badly. If you're a pixel geek that doesn't like your images twiddled with and you've painstakingly created your icon in Photoshop, you need to know the native size. If anyone finds a perfect image size (which may well not be square), please leave a comment.

SunSpider Benchmarks: WebKit Rocks

marcus@bointon.com (Marcus Bointon) — Wed, 19 Dec 2007 11:58:56 +0000

The WebKit guys have put together a new Javascript benchmark under the name "SunSpider". It's intended to go further than simple benchmarks like Celtic Kane's and try to emulate real-world tasks. Safari/WebKit has been getting pretty quick on these benchmarks anyway, but this new one really shows its strengths. There are various comments about people's results in the comments for that post, but no compilation for easy comparison, so I've put one together.
Updated: added Webkit Win and Opera 9.5b Win
Updated: Failed to run completely on Opera 9.5b Mac
Updated: Some stats for Opera 9.5b Mac and IE6
Updated March 18th: Added Safari 3.1, FF3b4, IE8

Continue reading "SunSpider Benchmarks: WebKit Rocks"

The Email Standards Project

marcus@bointon.com (Marcus Bointon) — Fri, 07 Dec 2007 10:02:17 +0000

The Email Standards Project is a worthy effort to try to get email clients to handle HTML email in a consistent way. Many already do pretty well, but there are some big exceptions: Outlook 2007 (with its ancient Word rendering engine), GMail, .Mac, Hotmail and others. Many are opposed to the whole idea of HTML email, but often their resentment is based on the fact that historically email client support has been so bad that they've had very poor experiences. Worse is that some senders (not us!) send HTML-only messages, which is certainly something that will drive a Mutt user potty. Smartmessages supports sending in plain, html and mixed formats (settable by each individual subscriber), and we ensure that our users get a clean, reliable platform for delivering their creations, so we try to work around the deficiencies of things like MS Exchange.

Generally the poor support in big-name clients has led to a need to develop HTML for email for very much the lowest common denominator, which for the most part means no CSS (unless you're prepared to tiptoe through the minefields of using it), no images, no scripts, no forms, no attachments. Too many designers think of email as being just like the web, but it's not - the vast majority of web pages will simply not work as email. These days the only effective way of designing for email is to start out with classic HTML 4.0 with no CSS or images and make your message look good using only type, white space and colour, because this is probably all that 90% (yes, really that much) of your recipients are going to see. You can then sprinkle a few images in for enhancement, but you should have no text in images that is not shown as text. With the advent of Outlook 2007's big step backwards, it's no longer possible to use background images, so you can't have text over images at all. You also can't rely on alt attributes as image fallbacks, as some big clients don't display the alt text if images are being suppressed as an anti-spam measure.

Many designers get very uppity about this kind of thing as it means that their palette of options is severely constrained, however, it should really be regarded as a challenge. It's not too hard to make stuff look good with heavy use of images (see CSS Zen Garden for gorgeous examples), but producing stuff that looks good with no images or CSS (or more to the point to still look good when those parts have been ripped out) takes a great deal of skill, experience and appreciation of the medium.

Any effort to try and raise the bar gets our support, so props to the Email Standards Project and to Freshview for starting it.

123 Strikes Again

marcus@bointon.com (Marcus Bointon) — Mon, 19 Nov 2007 09:20:25 +0000

You may have noticed that some of our sites have not been responding for the last few days. This is because 123-reg.co.uk had a name server outage. They didn't tell anybody or apologise at all, they just decided that several thousand people could do without their sites for a couple of days. People are generally pretty upset about it - just check out these damning blog entries.

123 have always been pretty useless, but to date I've not found anyone offering a decent professional service that also covered .co.uk domains. 123's big feature is that they are extremely cheap, unfortunately in every sense. This low price means that many of our customers have registered domains on there that we have ended up managing, so we have inherited their choice of registrar and default DNS host.
123 have NEVER responded to my requests for support, and I've reported major problems with their web interface many times - despite their takeover by pipex, their web interface has not changed at all (though the shiny home page has). It's not possible to log in to more than one account (something we need to do often) as their authentication system is totally useless - it's also impossible to log out (yes there is a link, but it doesn't actually do anything)! At least there are Firefox plugins to work around their ineptitude.

When transferring domains to 123, it's not possible to set up the DNS before the transfer has completed (or for them to simply retain existing name server settings - they always reset them to theirs), so it's impossible to transfer a domain to them without downtime and exposure of a nasty parking page on your domain.

Their "managed" hosting service is nothing of the sort. Steer well clear. I blogged about that quite a while ago.

All this adds up to something that is a lot less than professional. So from now on we'll be hosting our domains elsewhere, and suggesting that all our customers do the same.

I'm very happy to see that one of the better registrars I've used has finally got .co.uk accreditation.
They have a pretty and functional web interface, full access to zone files (if you want it), and they've answered every support request within a couple of hours (and with a certain Gallic charm). I've also had good experiences with enom.com, though while they are relatively expensive for uk domains, they have a UK support line that's not premium rate and is actually staffed by people who can do something about your request! The aforementioned blog post mentions Everydns, which looks like something to bear in mind if price is a real issue.

Google Carbon Footprint app launch

marcus@bointon.com (Marcus Bointon) — Wed, 31 Oct 2007 11:12:46 +0000

Over the last year I've been involved with the guys at d::gen. d::gen have put together the AMEE (Avoiding Mass Extinction Engine) Carbon Calculator, which has since been chosen by DEFRA as the official carbon calculator for the UK, and provides back end for the ActOnCO2 site as well as providing a public repository of official carbon emissions data.
Today marks the launch of the next big thing in AMEE's short history: Google's Carbon Footprint application, which is available as a gadget on Google's UK iGoogle home page.
The app was developed by Avenue A / Razorfish. My role at d::gen has been to deal with server and application configuration, deployment, hosting and monitoring, database configuration and load testing.
AMEE continues to grow in flexibility, ability, capacity and content, all while remaining a shining example of the 'right way' of running an open-source project.
Anyway, congratulations d::gen and AMEE, and thanks to Google and Razorfish for using us!
Now if you'll excuse me, I have to go and deal with the prospect of being on the receiving end of a link from a Google home page....

Microsoft wants more spam

marcus@bointon.com (Marcus Bointon) — Tue, 30 Oct 2007 09:45:28 +0000

If you read some of the postmaster docs on Hotmail/Windows Live, you might get the impression that Microsoft takes the processing of unwanted email seriously. However, this seems to be untrue.
Continue reading "Microsoft wants more spam"

Have I told you today how much Mac Minis rock as servers?

marcus@bointon.com (Marcus Bointon) — Fri, 05 Oct 2007 18:12:38 +0100

We're using a cluster of Mac Minis for hosting. They're running Ubuntu very nicely. I've gone on about them before (maybe not on here), but we just got a couple of new ones that use Core2 Duo rather than the older Core Duo models, and wow are they good. How good? Well, for CPU intensive stuff, we're seeing (per core) over double the performance of our Dell PowerEdge 1850 which sports two dual-core 3.0GHz Xeons (and these are the slower 1.8GHz models!). That's not bad for <20% of the cost to buy. They are unfeasibly small and really quite cute, plus they only pull about 120W (peak - 65W at idle!), so hosting them is dirt cheap (It'll cost you less than empty rack space - just ask Mythic Beasts whom we highly recommend, and they have minor niggles of the platform under control). Sure the disks are not too fast, and there's no internal RAID, but who needs RAID when you can have RAIC instead?

Munin per-plugin timeouts

marcus@bointon.com (Marcus Bointon) — Wed, 27 Jun 2007 22:06:01 +0100

I had a munin plugin that suddenly stopped reporting, and in the munin-node log there were lots of errors like this:

2007/06/27-06:50:26 Plugin timeout: myplugin : Interrupted system call

So it looks like the plugin is taking too long, but how to extend the timeout? Seems I'm not alone in wondering this. So I had a rummage in the source of munin-node, and it seems to want a 'timeout' parameter in its config. So I added a simple timeout 60 to munin-node.conf and it worked! However, that's a global setting rather than a per-plugin timeout, so I took a flying guess at putting the same directive in my plugin spec in /etc/munin/plugin-conf.d/munin-node like this:

[myplugin] timeout 60 user root

And that worked too!

Hackday

marcus@bointon.com (Marcus Bointon) — Mon, 18 Jun 2007 21:14:57 +0100

Well, Hackday was certainly an experience - the lightning strike and the roof vents opening was really quite disturbing! The lack of WiFi early on was a real pain as it made it very hard to connect projects, people and skills, so I suspect that most hacks were done by people who had arranged something beforehand, or who happened to meet at the bus stop.

Anyway, I got involved, along with Rob, Andy, Edward and James in something that turned into Fruitr - the search engine for unimaginative chefs. The basic idea was: take a picture of a piece of fruit, submit it to our site, and you're rewarded with recipes that use that fruit. Rob and James worked on using a C++ image recognition library (not sure which one) from a Pylons Python app, which was driven by statistical matching against images from Flickr that are tagged with 'fruitr'. Edward provided API access to his database of recipes courtesy of his site BigMunch, and I developed the testing web interface (later rolled into the main python app), and the email (in and out) interface. Andy had an abortive attempt at creating a J2ME app to talk to it, but was hampered by BlueTooth interference, buggy phones and poor mobile reception. We really had it all running by about midnight, and just a final polish was needed in the morning.

I spent much of Sunday looking at APIs to use with AMEE, without much luck or inspiration, then messed about creating an Apple Quartz Composer file that can be used as a screen saver. It grabs images tagged with 'hackdaylondon' from Flickr and presents them on faces of a bunch of rotating cubes. One face of the cubes displays live video (assuming you're on a Mac Book? or similar which has a camera). There is also a spectrum-driven audio bar display, and a rotating Hack Day logo, which is also moved by sound. You can download the Quartz composer file.

I did the presentations for both hacks. For the Fruitr one I just ran out of time - sweaty fingers and trackpads really don't mix. Because demoing a screen saver is not too hard, I dressed it up as a load of presentation faux-pas. I hope nobody thought it was for real!

Generally it was all pretty good fun, and it certainly had an excellent atmosphere, not to mention the endless supply of chocolate. Playing doubles tennis on the Wii at 2am was another highlight, along with the "non-showing" of Doctor Who...

I needed to leave before prizes and the Rumble Strips gig, so I missed out on that bit, and at the end of it all, I completely failed to grab a pink bean bag for Z. Oh well, I'll be back next year!

I'm going to hackday!

marcus@bointon.com (Marcus Bointon) — Sat, 02 Jun 2007 02:00:56 +0100

I got my invite for hackday today.

Anyone else coming?

3Ware RAID rebuilding

marcus@bointon.com (Marcus Bointon) — Mon, 14 May 2007 23:09:35 +0100

I've had the dubious honour of seeing some RAID failures and rebuilds lately. It's the kind of thing that doesn't get written about in the manuals very well, in particular what your RAID will report when it's having trouble. So, here are a couple of examples from a 3Ware RAID controller using tw_cli software. This is what tw_cli /c4 show displays when we have a dead drive:



Unit  UnitType  Status         %RCmpl  %V/I/M  Stripe  Size(GB)  Cache  AVrfy

------------------------------------------------------------------------------

u0    RAID-1    DEGRADED       -       -       -       149.05    ON     -      



Port   Status           Unit   Size        Blocks        Serial

---------------------------------------------------------------

p0     OK               u0     149.05 GB   312581808     G2109NHG            

p1     DEGRADED         u0     149.05 GB   312581808     G20X1BWG

So, we swap the drive, and it looks like this while rebuilding:



Unit  UnitType  Status         %RCmpl  %V/I/M  Stripe  Size(GB)  Cache  AVrfy

------------------------------------------------------------------------------

u0    RAID-1    REBUILDING     89      -       -       149.05    ON     -



Port   Status           Unit   Size        Blocks        Serial

---------------------------------------------------------------

p0     OK               u0     149.05 GB   312581808     G2109NHG

p1     DEGRADED         u0     149.05 GB   312581808     G209Y0HG

and after a little while...



Unit  UnitType  Status         %RCmpl  %V/I/M  Stripe  Size(GB)  Cache  AVrfy

------------------------------------------------------------------------------

u0    RAID-1    OK             -       -       -       149.05    ON     -



Port   Status           Unit   Size        Blocks        Serial

---------------------------------------------------------------

p0     OK               u0     149.05 GB   312581808     G2109NHG

p1     OK               u0     149.05 GB   312581808     G209Y0HG

There are plenty of obvious strings to match in this output (though there are many other reports available), so it's a reasonable thing to base a monitoring script on.

It's nice to see it actually work, and makes me extremely grateful that I bothered getting RAID n the first place. This would be a much unhappier post if I hadn't.

Dell RAID firmware and lockfile on Ubuntu

marcus@bointon.com (Marcus Bointon) — Sat, 28 Apr 2007 21:42:53 +0100

Ian P. Christian ran into this problem a while ago:

On a seperate note, anyone know how to upgrade firmware using Dell's software
on a non-RH system?

# ./RAID_FRMW_LX_R107404.BIN
/root
which: no lockfile in
(/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin:/opt/bin:/usr/i686-pc-linux-gnu/gcc-bin/3.3.6)
spsetup.sh: Cannot find utilities on the system to execute package.
Make sure the following utilities are in the path: sed lockfile tail rm mkdir
chmod ls basename
/root

'lockfile' is missing - whatever that is!

Lockfile just doesn't seem to exist outside of RedHat (e.g. lockfile-progs on Debian doesn't include it), however, you can of course find it on a RedHat system, and I happen to have one handy. I just copied the binary to my Ubuntu installation, where it appeared to run just fine, and allowed the firmware updaters to run ok. Thought someone might like to know.