New Zealand is a long way from the UK, but that doesn’t matter once you’re here (and it’s worth the trip). The scale is quite different from the UK (bigger country, fewer people) and the attitude is sometimes described at ‘number 9 baling wire culture’. As in whatever the problem, you can solve it with ingenuity, lateral thinking and a twist of baling wire.
That applies across the business scale. The door closers in the restaurant we ate at tonight (the excellent White House in Wanaka, notable both for the delicious food and the fact that the by-the-glass price for wine is the same as the by-the-bottle price instead of nearly twice the price) are spaced to the correct distance so they close gently rather than slamming - with preserving jar lids. Researchers in Wellington have been getting excellent bandwidth with wokfi: cooking implements like woks and chinese strainers used as WiFi antennas (because they’re the right shape). On a larger scale one of the major breweries here, Monteith’s, needed a new depalletiser, to take bottles out of the pallets for filling. Instead of paying NZ$250,000 for a standard model, the brewers recycled two old vans and a handful of other parts to create a custom machine that does exactly what they need for a fraction of the price. Like custom software, it needs maintaining but the original developer is right there today (turning beer barrels into a rather stylish set of sinks and urinals).
There are plenty of developers in New Zealand who have embraced Web 2.0 ideas, not because they’re trendy or because it’s a shortcut to VC funding, but because it works for what they want to do. Software on demand startups make the business pages of the local paper when they get a contract from the district council. A combination of pragmatism, ability and a can-do attitude is the typical Kiwi approach. That’s balanced by a national dislike of people pushing themselves too far forward: ‘tall poppies’ are likely to get scythed down. And if you get it wrong there’s no easy ride: at a recent company meeting, shareholders of one business savaged the independant directors for championing an unsuccessful deal. But again, it’s a hands on approach to things.
-Mary Branscombe

P.s. If you can read this post, add another to the list of things you can do from a BlackBerry halfway up a mountain.

Your BlackBerry is also a reasonable if slightly limited system administration tool when you’re on the road. We discovered this the hard way when the DSL line connecting us to our mail server back home in London failed. Even from New Zealand, we knew the mail server wasn’t there any more. For one thing, the flood of email had stopped. For another, the BlackBerry Internet Service that T-Mobile uses sent an error message saying it couldn’t reach our Exchange server. But was it a power cut, a server crash, a problem with the router or a problem with DSL? If it’s a power cut or a crash, we have to phone the downstairs neighbour (we really must put a battery in the UPS). If it’s the DSL, we only need to wait 24 hours for the mains timer to power cycle the router and restore the connection. It’s possibly work at the exchange, it could be fluctuations on the line; we don’t know what causes it to happen, but if the DSL connection is interrupted, power cycling the Zyxel Prestige firewalling router we use seems to be the only way to see the outside world again.

A BlackBerry makes an excellent diagnostic tool. Download a copy of the free MidpSSH (http://www.xk72.com/midpssh - which is also available for most other Java-enabled smartphones)  for a secure telnet client that can work over SSH connections. Once you get to grips with switching between input and result screens you can work with most telnet-based management tools with relative ease. Zyxel’s management platform includes a set of system diagnostics that let you ping systems inside the firewall, so once you’re connected to a router over telnet or SSH, you can check to see if failed systems are responsive. Of course, if you can’t connect, then there’s a problem with the line.

It’s a good idea to keep a web-based network diagnostic tool in your bookmarks. Just fill in the name or IP address of your firewall, and use ping to see if it’s online. Traceroute will help show if a failure is at your router or inside your ISP (just make sure you’ve memorised the name of your ISP’s DSLAM or gateway server, so you know where the problem is, and whether BT’s to blame…).

There are other network tools for BlackBerries, but the old school approach is still worth using. If you’re telnetting to an application-specific port (like POP3 or IMAP) and you’re feeling rusty, you can use the BlackBerry browser and Google to show you the commands you need - and the portsDB site(www.portsbd.org) to show you which ports are used by which service…
In our case, we couldn’t make a telnet connection to the router’s management interface. A quick traceroute later, and we knew the problem was the router not recovering cleanly from a DSL failure. So all we had to do was wait for the timer to power cycle the router. 17 hours later mail began to flow again. While we’re on the other side of the world, a GPRS connection and a BlackBerry mean we’re still in control of our network.
- Simon Bisson

Walking round the Dreamforce expo hall, we saw several companies bringing the identity metasystem to Salesforce.

Salesforce uses delegated authority authentication so you can have your users log in with their standard account system - or you can replace it with your own authentication system. If Salesforce isn’t the only system you use, then it’s yet another set of accounts and passwords to manage - to hand out to new users, to store in case users forget them and to deprovision when people leave. So the option to work with your existing tools will be welcome to larger companies.
Iovation has an identity tool for Salesforce, that lets you log in using your OpenID. Sxip (Simple eXtensible Identity Protocol, pronounced Skip rather than Sexip if you were wondering, or indeed Ship if you’ve spent as much time in Portugal as Simon has) has an appliance that gives you single sign on for multiple systems and you can add Salesforce as just another system.

Sxip also has an open source tool called the Whobar (www.whobar.org) that lets you add multiple identity systems to any Web site. Instead of coding up a chunk of Perl yourself you run the Whobar and as well as your existing username and password system you’ll be able to accept OpenID, InfoCard and i-names. This is a very simple way of enabling the new identity systems, so you can afford to try them out to see if your users want them.

 InfoCards will be easy to get once Vista and Internet Explorer 7 finally ship and if Microsoft keeps its promise of adding InfoCard support to Windows Live ID (the next generation of Passport) they’ll become widespread pretty quickly. A lot of people already have an OpenID without knowing it; if you use LiveJournal you’ve got one. Wikipedia will switch to OpenID when version 2 comes along. If not, you can get one free from OpenID or Verisign. Today there’s not a lot you can do with it beyond using a few wikis and photo hosting sites or leaving comments on a LiveJournal blog without getting an account. But as the identity metasystem starts to become a reality, OpenID will be a handy way to log in to multiple Web sites without having to remember dozens of passwords (or putting convenience ahead of security by using the same password on every site).

So when we arrived in New Zealand (note: a three hour layover is infinitely more pleasant at Auckland airport than at LAX) this week and discovered that Eurekster is planning to allow users to contribute information as well as links to its swickis, we suggested they look at something like OpenID. Swickis are pre-tuned group searches around particular topics; if your Web site is about New Zealand wines then you and others with the same interest can refine general search results so visitors searching for ‘Marlborough’ get the South Island wine areas rather than the British private school. And links in results pages that other visitors have clicked on and spent time at get a higher ranking next time.

Swicki comes from ’search wiki’ and so far the emphasis has been on the search side; letting users contribute information adds more of the wiki idea. The point of having a Eurekster identity will be so that visitors can see who has contributed a particular nugget; it’s a reputation system. Using an existing ID system would let you see the reputation of that same person on Wikipedia - and maybe other systems in future. You’d trust the sky-diving ‘expert’ on a swicki more if you could see they hadn’t just bought their first parachute on eBay the week before…

We have had some dreadful arguments over the years. We still disagree over how the Windows XP memory manager works and we can’t agree on moving My Documents to the server versus offline files. The worst was a three-hour stand-up knock-down fight over digital rights management (damning us to a digital dark ages that will wipe out a generation of content for ever versus a rich permission system that enables more flexible publishing by automating rights and requirements? You decide). But the trip when we argued from Birmingham to Leeds about whether every application is fundamentally a database sticks in my mind too.

My thinking here is that applications access, manipulate and store data. Editing an image? Data. Creating a presentation? Data. Saving high scores in a game? Data - and the storyline, rendering information and all the rest of the game is data to start with too. What about the algorithms and the image editing filters and the transition effects in the presentation? You can think of them as stored procedures and triggers.

That’s what the new Salesforce programming language Apex gives you, according to senior VP Steve Fisher. “We’re modelling Apex on triggers and stored procedures and things that make databases effective. When you work at a basic database level you create, delete, update objects - all we’re really doing is running a pre update and  post update trigger, a pre or post save trigger or a pre create or post create trigger that jumps in before the usual code can run”. Salesforce’s own code is a mix of SQL and Java but what it manipulates is an Oracle database - and all the applications on Salesforce run on top of that database.

There was a lengthy discussion after the initial Apex announcement about whether Salesforce is going to buy and build more applications beyond their core competence of CRM. They already bought a tool for monitoring Google adwords because it proved so popular with users. Will they buy or build more applications and get bogged down by them?

Personally, I think competition with Salesforce partners might be more of a problem. Partners can build features that users are asking for and Salesforce suggests they might do it before an official release comes out. But what happens to the partner when Salesforce catches up and the code they invested time in building  is available for free? That’s the balancing act Microsoft and Apple already have to make and when their feet slip on the rope partners like Symantec get unhappy. Can Salesforce juggle the same demands or will they back away from applications  - maybe even the original CRM in time - and stick to the platform?

So Salesforce today - platform or application? Salesforce the company believes Salesforce the software is a platform, builds on it as a platform and sells it to users as a platform. But there’s a measure I use to tell developer and user conferences apart; there are still far fewer women at developer conferences. I haven’t had to queue for the loo since Oracle OpenWorld last summer but the queue for the restroom after the first Dreamforce keynote stretched out into the corridor. To the users, Salesforce is still an application.

Mark Benioff, CEO of Salesforce.com has just published a book on corporate responsibility, The Business of Changing the World; timely given the investigations of HP and Apple that have been front page news here in California over the last few days. My view on corporate responsibility; it would be nice to see some.

Benioff also invited General Colin Powell to give a keynote speech at Dreamforce (the Salesforce conference we’re attending this week) because Powell’s work inspired Benioff to set up the Salesforce Foundation. He had some serious comments about the post Cold War world changing from one known battlefield to a series of playing fields where economics, energy, the environment and education matter more and more. Democracy won’t get very far without the economic benefits the US claims it always brings.

But he also has a good line in repartee, from joking with Gorbachev that he didn’t want to give up Russia as an enemy if it meant giving up his budget to coming abruptly down to earth the first time he went back to flying on a commercial plane like the rest of us. He didn’t have any luggage to check, he’d paid for his ticket in cash and he turned up late for the flight; naturally he was pulled aside for a secondary security check. And as they were scanning his shoes and turning out the nooks and crannies of his carry-on, the security chap looked up and said ‘welcome to the airport, General Powell’. And Powell asked him ‘If you know who I am, why are you searching me?’. He didn’t get an answer but if he had, I’ll bet it would have involved the words ‘the system’ or ‘that’s just the way we do things’.

There’s identifying someone and there’s knowing what to do about who they are. The first part of that is technology; the second is business process. One of the advantages of Salesforce is that you don’t need to get the IT team involved so the people who work in the business can create a system that reflects the way they work. Lightweight technology can be an easier way to model a business process. The downside would be if you need up with too many business processes in Salesforce and not enough tied back to your audited, compliant enterprise system where all the reports are generated and the metrics calculated.

In many ways, Salesforce is the Excel or the FileMaker of the on-demand world; you can knock up some handy tools and models quickly and get a lot of work done with them. But how do you keep tabs on that and integrate it with the rest of your technology? Do you dump your other systems and move everything online or - more likely - look at integrating Saleforce into your existing systems?

And the technology of identity? More on that after the frighteningly early breakfast meeting I have planned for tomorrow with Sxip.

Mark Benioff, CEO of Salesforce.com has just published a book on corporate responsibility, The Business of Changing the World; timely given the investigations of HP and Apple that have been front page news here in California over the last few days. My view on corporate responsibility; it would be nice to see some.

Benioff also invited General Colin Powell to give a keynote speech at Dreamforce (the Salesforce conference we’re attending this week) because Powell’s work inspired Benioff to set up the Salesforce Foundation. He had some serious comments about the post Cold War world changing from one known battlefield to a series of playing fields where economics, energy, the environment and education matter more and more. Democracy won’t get very far without the economic benefits the US claims it always brings.

But he also has a good line in repartee, from joking with Gorbachev that he didn’t want to give up Russia as an enemy if it meant giving up his budget to coming abruptly down to earth the first time he went back to flying on a commercial plane like the rest of us. He didn’t have any luggage to check, he’d paid for his ticket in cash and he turned up late for the flight; naturally he was pulled aside for a secondary security check. And as they were scanning his shoes and turning out the nooks and crannies of his carry-on, the security chap looked up and said ‘welcome to the airport, General Powell’. And Powell asked him ‘If you know who I am, why are you searching me?’. He didn’t get an answer but if he had, I’ll bet it would have involved the words ‘the system’ or ‘that’s just the way we do things’.

There’s identifying someone and there’s knowing what to do about who they are. The first part of that is technology; the second is business process. One of the advantages of Salesforce is that you don’t need to get the IT team involved so the people who work in the business can create a system that reflects the way they work. Lightweight technology can be an easier way to model a business process. The downside would be if you need up with too many business processes in Salesforce and not enough tied back to your audited, compliant enterprise system where all the reports are generated and the metrics calculated.

In many ways, Salesforce is the Excel or the FileMaker of the on-demand world; you can knock up some handy tools and models quickly and get a lot of work done with them. But how do you keep tabs on that and integrate it with the rest of your technology? Do you dump your other systems and move everything online or - more likely - look at integrating Saleforce into your existing systems?

And the technology of identity? More on that after the frighteningly early breakfast meeting I have planned for tomorrow with Sxip.

We try out a lot of new mobile devices; just keeping up with HTC means changing phones five or six times a year. Many of them are Windows Mobile devices and once we’ve gone through the intricacies of installing our self-certified (Do not pay Verizon $150 dollars, do not pass Go) root certificate, email starts flowing from the Exchange server.

It’s at this point that things get noisy (painfully so if you have a headset plugged in). The email notification shrieks, buzzes, waves and flashes like a toddler who’s had too much tartrazine. 168 emails! 168 emails! Read them while they’re hot! This attention-seeking is why mobile email gets a bad name. I want email to arrive immediately but I’ll look at it when I’m good and ready. I may want to know when a new message arrives if I’m waiting for something urgent but I’ll usually wait and check when I have time to deal with messages. And a discreet vibration in my pocket is quite enough; I don’t need an on-screen alert and a sound and a flashing LED as well, which is the usual default for Pocket PC phones (and on everything except the Palm Treo 750V the notification camps on top of the Contacts soft button and won’t go away).

I might want that kind of alert if the hard drive on the server was about to melt down or the rack temperature was up to gas mark nine, especially when we’re on the road and not looking at more general logs and reports. We automatically power cycle the DSL router every day at 3am when we’re travelling so that if BT does work at the exchange and the DSL connection needs restoring, we’re at most 23 hours and 59 minutes away from an automatic fix and I’m considering a dead man’s switch alert from another system if that doesn’t complete correctly. But I don’t want to see everything and not many mobile devices offer different alerts for the messages I consider a priority (though I can have any ring I want to warn me it’s an editor calling rather than my sister).

Over a decade ago a friend of mine wrote a monitoring and alerting package for servers that had a then-revolutionary feature; as well as updating a log file it could send an email or a page for particularly important alerts. Written for Unix ,it proved hard to port to Windows NT (Services for Unix would make it far simpler now), and he didn’t implement some of the more advanced plans for notifications. Now things are tending the other way.

Discussing the next generation data centre with Olivier Helleboid, VP of adaptive infrastructure for HP’s Technology Solutions Group this week, he suggested that the automated light-out data centre is going to have far fewer alerts to deal with; everything will be logged but a lot of problems can be taken care of automatically.

You do need to know about the things that need manual intervention and many server management tools can email or SMS you these days. What you can’t yet do is set up rules on the server that tell your mobile phone when the alert is something to make a song and dance about.

Two main topics have kept cropping up in our meetings with hardware and software companies alike over the last fortnight; the move to parallel execution as multiple cores become common, and machine learning making its mark in the real world.

Back in the day, as they say over here, I did a degree in artificial intelligence and knowledge-based systems. Over the last 15 years I’ve seen more and more of the techniques we looked at move into the mainstream; AI bots are common in games and on IM networks, speech recognition and handwriting recognition aren’t solved problems but you can use them to get things done in limited ways without people thinking you’ve walked out of a science fiction story, Bayesian statistical models protect you from spam quite successfully and what we’ve learned about human cognition guides principles of interface design. And we have a robot that scrubs the kitchen floor at the push of a button.

At the time there were two different directions for handling knowledge. You could interview people and codify both their knowledge and hunches as rules and facts to build an expert system, or you could analyse data to build those rules without human intervention. Neural networks learned rules that you couldn’t translate into English because they’re basically mathematical weightings for numerical transformations and equations that approximate the behaviour of a data set. But machine learning often combines human knowledge and data analysis now; you can give the system hints to get it going or to improve accuracy on the data you train it on to get better results for the data you want to work with.

So far this month we’ve seen Intel investigating this to work out when to power down a PC and when to turn it back on far more flexibly than just setting timeouts for different components; the right rules could give you an extra 25% battery life. Mark Najork at Microsoft Research in Silicon Valley is using machine learning to learn characteristics of spam Web pages; Dennis Fetterly and Steve Chien in the same group have built a classifier that learns what characterises a blog page. Machine learning has gone from a research topic to a real programming technique and more cores could make it more common.

Multicore processors mean you need several things going on at once to realise the potential of the processor. Your operating system and the other applications running can use the other cores if your main application doesn’t need them, but if you want to speed up a single application you need multiple threads. Intel is tackling that at the compiler level as well as producing parellelised versions of common algorithms. We need those. When I did my degree parallel programming was an advanced research topic many people thought might never make it to mainstream processors and parallel code was frightningly complex; only one of those has changed. Game developers can use a whole core for a physics engine to improve the realism of a game. Maybe as CPUs get more and more cores, applications will start to use one of them to for learning. Spelling corrections, messages you mark as spam, links in search results you do and don’t click on; your system could really start to customise itself for the way you work.

Two main topics have kept cropping up in our meetings with hardware and software companies alike over the last fortnight; the move to parallel execution as multiple cores become common, and machine learning making its mark in the real world.

Back in the day, as they say over here, I did a degree in artificial intelligence and knowledge-based systems. Over the last 15 years I’ve seen more and more of the techniques we looked at move into the mainstream; AI bots are common in games and on IM networks, speech recognition and handwriting recognition aren’t solved problems but you can use them to get things done in limited ways without people thinking you’ve walked out of a science fiction story, Bayesian statistical models protect you from spam quite successfully and what we’ve learned about human cognition guides principles of interface design. And we have a robot that scrubs the kitchen floor at the push of a button. At the time there were two different directions for handling knowledge. You could interview people and codify both their knowledge and hunches as rules and facts to build an expert system, or you could analyse data to build those rules without human intervention. Neural networks learned rules that you couldn’t translate into English because they’re basically mathematical weightings for numerical transformations and equations that approximate the behaviour of a data set. But machine learning often combines human knowledge and data analysis now; you can give the system hints to get it going or to improve accuracy on the data you train it on to get better results for the data you want to work with. So far this month we’ve seen Intel investigating this to work out when to power down a PC and when to turn it back on far more flexibly than just setting timeouts for different components; the right rules could give you an extra 25% battery life. Mark Najork at Microsoft Research in Silicon Valley is using machine learning to learn characteristics of spam Web pages; Dennis Fetterly and Steve Chien in the same group have built a classifier that learns what characterises a blog page. Machine learning has gone from a research topic to a real programming technique and more cores could make it more common. Multicore processors mean you need several things going on at once to realise the potential of the processor. Your operating system and the other applications running can use the other cores if your main application doesn’t need them, but if you want to speed up a single application you need multiple threads. Intel is tackling that at the compiler level as well as producing parellelised versions of common algorithms. We need those. When I did my degree parallel programming was an advanced research topic many people thought might never make it to mainstream processors and parallel code was frightningly complex; only one of those has changed. Game developers can use a whole core for a physics engine to improve the realism of a game. Maybe as CPUs get more and more cores, applications will start to use one of them to for learning. Spelling corrections, messages you mark as spam, links in search results you do and don’t click on; your system could really start to customise itself for the way you work.