Brick Lane is basically "Little Bangladesh", a road in the East End of London which is jam packed with Indian restaurants, mostly specializing in Balti dishes. If you dig Indian food, you absolutely must try it. When my Dad, Charlie and Denise all lived in London we had the goal of eating in every restaurant on Brick Lane, but lost track.

Another great feature of Brick Lane is the 24-hour Bagel shop at one end of the road, a legacy of the days when this was a Jewish neighborhood (before which it was a Hugenot neighborhood). Actually there are two bagel shops, but one is almost always deserted while the other is always packed. They bake bagels and pastries in the shop, and you can buy them for 10 pence each.

Anyway, the last time we visited Brick Lane we were annoyed by touts, who have become more agressive and numerous since "our day". Apparently the local council is concerned that the touts are getting out of hand to the point where it may turn people off of Brick Lane and hurt business. It would be nice if the restaurant owners realized what a turn-off the touts are and put a stop to it themselves.

Via the Big Smoker

I've added a new book to my shortlist of bibles for Systems Administrators. The Purple Book is another must-have for Unix admins, as is the Unix Power Tools book. But this one isn't just for Unix admins, it applies to anybody managing multiple systems.

It's a meaty book, with a satisfying heft that isn't fluffed out with purty pitchers or code listings. Although its approach is to avoid being platform-specific, so as to appeal to both Unix and Windows (and probably miscellaneous), it isn't wimpy. The authors cover nuts and bolts of why you should buy proper server hardware rather than put your exchange server on a desktop machine, stick it in a rack with proper rack-mount rails rather than on a shelf, etc. Meat and potatoes stuff.

The value of this book is that it does approach things from the "why" perspective, especially "why" from the point of view of someone who needs to manage a bunch of systems. It's the kind of thing missing from OS manuals, which tend to focus on "how". The fact that it's not vendor-specific means they can take a realistic attitude about selecting tools and processes, rather than just advocating whatever comes with the system and is recommended by the vendor.

I highly recommend this book for any sysadmin who is learning about the best way to manage a group of systems, which, let's face it, should be any sysadmin whatever their skill level. Check out the table of contents for a feel for what it covers.

So the makers of the software I use for this website, a company called SixApart, have released a beta of the new version, which is supposed to have been rewritten to make it easier for Perl-heads to modify. Unfortunately, they've also changed their licensing policy, which is potentially going to force many users who have gotten used to having it free to pay to upgrade.

It's funny, in the zillions of posts about the upgrade, I have yet to see a single person comment on what's changed technically. I would like to give it a spin, but probably won't have time for a while. But I'm curious, because one of MT's weaknesses has always been that it's fairly inflexible, so I'd like to hear how good a job they've done to improve that.

Personally, I'm not that burned about the new model. Partly it's because I'm still entitled to use it for free - I would only need to pay if I had more than one person posting here. I think it would cost $69 if I wanted to use MT for a blog with up to 4 friends, which doesn't seem horrible. I do think they could have made the licensing a bit more liberal, personally I'm a fan of the model Dan Bricklin recently suggested for a small software company. A month or two ago I had pondered the best way to structure licensing if I ever get around to setting up my own company, and what I came up with is almost exactly what Dan suggests.

But even if it's not my personal preferred model, and does have some rough edges, I think SixApart have done a couple of very smart things with their new model. The best thing they've done is open things up for third parties to earn money from their software. Their license used to esentially forbid consultants and hosting companies from charging clients to set up and host MT.

This has always struck me as counter-productive. I guess the intent was to give SixApart a monopoly on consulting and hosting revenues, but doing that severely limited their ability to grow. Other than users who are technically literate and can install and support themselves, your revenues are limited by your staffing levels. If you let third parties make money from your software, they'll go out and sell copies to their clients, with little added cost to you.

Movable Type is a good piece of software. The static publishing model is one that should be a no-brainer best practice which dramatically improves the scalability and reliability of a website. But very few other blogging or CMS applications use it. I hope the user community will get over its outrage, and that SixApart will smooth out its licensing and prosper for many years.

I've been at my current job for over a year and a half, and now I'm moving on. I love the company I work at now, but my role within it has become more and more mired in IT management, as opposed to developing products and services. It's certainly been a useful experience, I've gotten to manage the IT budget, wrestle with building an infrastructure to manage loads of systems, and play with different server technologies. I've strengthened my knowledge of OS's (especially Linux) and networking, but it's time for a career course correction.

My new job is Java development, which is great. I love Java programming, but lately for me it's dwindled to occasional tinkering with our in-house build management system and some personal projects, and getting vicarious thrills from reading Java Blogs. Getting back into a fulltime development role is what I really need, otherwise I'll soon become unemployable. Already it was tough getting through the screening of recruiters with only a dim understanding of technology, who look at my current job and don't see the right keywords.

Ironically, I had several experiences of recruiters not forwarding my CV to a company until several weeks later (i.e. after they've tried out candidates they thought had a "better fit"), but then I did very well once I got through. People actually working on Java based projects can see the value of someone with a good development background who knows application servers very well.

So what's the new gig? It's a startup company, well-funded by VC's, which builds and runs mobile phone services (e-commerce type stuff) on a Java-server platform, mostly SMS and WAP. I like it because it's heavy Java server side (Weblogic, Unix, Oracle) development, and also involves managing the application, so I still get to do the service management. I especially like the fact that I'll be working on applications owned by my company, rather than building or managing something for clients - the dynamic is just different, and in my experience more satisfying.

I also have a feeling of timewarp, going to a small startup with VC funding, which is offering stock options and doubling its users every month. The other day I saw a grown man on one of those push-scooters, which I took as a sign. But this time around I think I'll stay away from the stock market.

So I reinstalled again, this time selecting the core software group and not monkeying with the defaults. Yesterday I had added a few things. It's still a crippled installation, no "man", no decent shell, etc. This would not be bad, after all a minimal install is a reasonable thing to expect when installing only the core packages.

But you'd like to have a fairly convenient way to start with that minimalist system and install new packages on top. How about a tool that lets you walk through the packages available on the installation CD, tagging ones you would like to install? Nope. How about letting me mount the CD so I can trudge through the packages by hand, reading description files (or using a references to the packages) and running pkgadd on the ones I'd like to add?

"No soup for you!" shouts the Solaris installer.

The core system install doesn't include the volume manager, so you have to mount the CD by hand. Not a horrible thing for the veteran Solaris admin, but I guess I'm not veteran enough. You see, the Solaris 9 installation CD is partitioned into multiple slices, which are the Solaris version of disk partitions, and the slices don't all use the standard CD file system format. The slice that mounts with the usual mount command isn't the one with the packages on it. There may be an incantation that will let me mount the slice with the packages without shouting "bad magic number! Bad, BAD magic number!", but I couldn't Google it out. So in the end I put the CD on a running Solaris box and ftped the volume manager packages to the new box. Thank hell the bare bones Solaris install at least includes an ftp client!

I feel like such a newbie, even though I've been managing Solaris boxen for at least 7 years now. I guess most of what I've been doing in the past few years is running existing boxes; my last install was a Solaris 7 box, back before the millenium rolled over.

Still, it shouldn't be this hard.

So I installed s2putty ssh client on my shiny new 6600 phone, but it doesn't appear to work with OpenSSH on Solaris. It connects to my server, but complains that the server won't enable compression. OpenSSH doesn't appear to support compression on Solaris, at least not while privilege separation is enabled, which is a pretty fundamental security measure.

Having the ability to show my geek friends an "ls" of my home area on my phone would be cool, but not cool enough to open a security hole on my server. I don't currently have any Linux boxes with ssh ports open outside a firewall, so that's the end of this for now.

So I finally managed to get Solaris 9 installed on an old sparc box here at the office. It turns out there was an option before the software selection for which I accepted the defaults, and which imposed a bunch of package dependencies outside the core software group. The problem is, I appear to have missed some important packages such as, oh, "man". So I'm going to have to comb through the available package to figure out what else belongs.

Once I get a decent, vanilla core system I will try to make a Flash installation CD, which can be used to clone the current setup. I have another spare Sparc box handy to test it on, although I'm going to have to dig up a power supply and some RAM since we have used the box several times as an organ bank for ailing development servers.

Another problem with the Solaris installer is that it did a poor job of picking up its network settings. Although I told it to use DHCP to configure itself, and the DHCP server shows it made some requests, the installer still prompted me for its domain name and DNS servers, which it then claimed were "incorrect". And when I booted the newly installed box, it had not picked up its hostname (and it never asked me for it), so it called itself "unknown" until I set it manually.

While I'm on the subject of Solaris, I have to say their command line installation is unforgivably poor. I've been working on installing Solaris 9 on a box, and migrating our half-dozen-plus disparate Solaris 7 boxes to an up to date, cfengine managed well-oiled infrastructure. It's a well-overdue upgrade, but the installation is so nightmarish that every time I start I give up and mvoe on to higher priority tasks.

They've got a GUI, webstart installation interface, but these are servers without graphics cards, so it does me no good. I could do a network install, but it sounds like I'd be best off having a Solaris 9 box as the install server, which means I need to get at least 1 box installed from the CD first.

The problem is the software installation. When you install Solaris you choose a software bundle, basically a user workstation with CDE (GUI interface), a developer workstation (user workstation plus Sun's dev tools), and a full version. But for a server, I just need the core.

During my attempt today, for some reason it wouldn't let me install the core, or anything other than the full whack, because it complained that some software packages depend on things not in core. It doesn't offer the option to leave those packages out, or to go back to an earlier stage to see there was something else I could have chosen.

So I had to select the full whack, and attempt to deselect the packages I didn't want. This is an excruciatingly painful process. There are at least 10-20 screenfuls of packages. For each packages, there are about 4 or 5 steps to go through to deselect it, each of which involves painfully slow re-scrolling of pages on the TTY console I'm using. I got through about 3 pages of packages before I accidentally hit one "enter" too many, and apparently told the installer I was done selecting software.

There's no option to go back.

That's it.

Start over.

I really hope Sun gets a clue before they swirl down the drain. I can live with their Java policy - it's free and comes with source, and I have no intention of modifying it and redistributing Kief's Super JVM any time soon, so the fact that it's not Free isn't a dealbreaker for me. But Solaris is what Sun is all about, and it's been lapped by Linux in terms of TCO.

Linux isn't just cheaper in software licensing and hardware costs, it's less hassle to manage on a daily basis. I can handle the bigger purchase price, since there are still plenty of big clients who are reassured by big price tags, but time spent monkeying with poorly designed management tools comes out of the time my team and I could be spent doing more productive things.

In the past year or two I've become a Debian Linux fan, in part due to the ease of keeping a Debian box's software up to date. Other systems, including Solaris, have package management, but apt-get lets you pull software from a desitributed repository, finding the latest versions and dependencies for you.

So every morning I get a cron-generated email listing the software packages that have updates, and a single script command grabs them for me. Installing a new package is a snap, I just run apt-get with -s to see what other packages will be installed, then run it again without -s to make it so.

Well it turns out some folks have created something called pkg-get which does the same thing for Solaris systems. Although it works with the SunFreeware site, that site doesn't use the newer package format that identifies dependency, so using blastwave.org is the best bet.

Blastwave lists 574 packages, which is peanuts compared with Debian, but a respectable core. It looks like it covers the essentials, but has some serious holes - they've got plenty of PHP, for example, but I only spotted a single Java package, Jetty. To be fair, I prefer to download most Java apps myself rather than use apt even on Debian, since the compiled distributions usually work fine and I prefer to use my own directory layouts. But a Tomcat package would be handy. If I get some spare time (hah!) maybe I'll have a go. First I need to find the time to try pkg-get on one of my own systems to see how I like it.

I'll vote for Kerry, but it's a shame to see him pandering to anti-trade sentiment. Rafe is like many who buy into Kerry's opportunistic taxing of overseas revenues:

taxing overseas revenue at a lower rate than domestic revenue is just stupid

Actually, taxing overseas revenue at all is something pretty much unique to the US. Few other countries get away with taxing money made in other countries, and it's a disadvantage for American companies. Kerry's proposal, popular though it may be among the types who consider lottery tickets to be a good investment, handicaps American companies selling products in foreign countries, since their competitors don't have to pay double taxes.

Not a bad thing for those wishing that American brands didn't dominate the global market, but certainly a bad thing for people worried about "hollowing out" the US economy.