People like having their expectations met and being treated as the person they believe themselves to be, even if that makes them miserable.

I wish people were more responsive to having their hopes met, or being treated as the person they WANT to be, but these tend not to work as well in practice. Most seem to put far more energy into reinforcing their current outlook and identity than pursuing a path that would lead them to actual happiness.

Weird that some people like not to be happy.

Adderlan @ High Performance on Wall Street (Fall 2008)

Flagg Management and the A-Team Group throw these nice little events every now and then which I have been attending with some regularity. Previously ones were called "XML on Wall Street", "Linux on Wall Street", and "SOA on Wall Street".

The theme of this show might as well have been "Low-Latency on Wall Street".

Now my primary reason for attending these events is to get a idea of how Wall Street views I.T., but what I usually end up with is an idea of how I.T. views Wall Street. Not as useful to me as the former, but still valuable. At the very least I can make some assessments on current technical trends.

All of the software products shown (such as grid computing, parallel programming libraries, low-latency messaging, data format specific sorts, and profiling tools) have open source or basic operating system feature equivalents, and I believe the same thing that happened to the Browser, IDE, and Virtual Machine market, will happen to these markets even more rapidly. That is, prices will be driven down to the point where the segment will largely cease to be profitable.

And I have this nagging suspicion that many of these products don't offer a legitimate technical advantage even today.

Latency has already been solved to the point where it is no longer a issue of optimization but of trade off, and the speed advantage of various messaging and sorting products is entirely due to being a good match for a specific set of tasks and data formats. When it's close, you can easily get multiple orders of magnitude improvement. It all depends on how closely it happens to match the current activities and architecture of your data center at the time they are used.

And regardless of how good the software is, it cannot overcome the inherent latency of the hardware itself, which has only been increasing with each new generation.

These are also the EASY problems. Matching a sort to a data structure is easy. Sending lots of tiny messages quickly is easy. Taking advantage of multi-core processors with virtual machines is easy. Profiling a system and showing the raw data is easy.

Being able to take full advantage of a multi-core system running arbitrary code is hard. Generating assessments based on profile data is hard. Calculating the optimal cache strategy is hard. Fighting the hardware trend of increasing latency is hard. Perhaps these problems are being left for the programmers, but the programmers that are able to address these can also address the easy problems too, and with more effective solutions than the products.

The hardware present was a lot easier to asses, as it's almost all commodity at this point. Even Cray is making x86 boxes now. The only deciding factors at this point are cost and support. Support is especially difficult to asses, as the only way to gage it legitimately is to invest time and money in the product to begin with, and it can get better|worse over time. Sure, you could also rely on recommendations and brand names, but those metrics are often far from accurate.

There is still a cognitive mismatch between how I.T. works in reality, and how it is presented for sale. Software is almost universally marketed in terms of discrete products that can be connected through a workflow at this point, but the real world doesn't work like that. There was considerable functional overlap between many of the products, if not outright redundancy, and integration is a question that can't be answered with just lines and boxes.

If you're going to one of these events to get some useful answers in addition to entering the raffles (I won a camera BTW, thanks Platform, but shame on you for suffering from the 'Flash hiding the HTML menu' problem on your website), bring a description of your data center, and a list of the ten most common tasks run on it, show it to each vendor, and simply ask them what they can do for you.

If at all possible, put your photo on your business cards. After the show I had over 40 cards, and I was having a hell of a time connecting them to the people I spoke to. Beyond your body language, your face is your non-verbal name, and you should use it.

I'd also go a bit beyond business cards, and bring a one sheet description of yourself or company. This is different from a resume in that it details what you are DOING, and not what you have DONE. Almost every vendor asked me about what I did and who I was with, and I had no simple answers. Business cards are also often thrown away after being used to add you to a mailing list, but a one page may stick around and get to places the card wouldn't.

I'll cover some of the more exceptional companies, such as the understated NetApp, and the enigmatic Enigmatec, in more detail in the future. You can also find me at upcoming Flagg Management / A-Team Group events through my calendar.