Exchanges used to be member-based organizations and paragons of capitalism. They were fully dedicated to stability, reliability and the principal that a market must never show weakness, never lose a message and never go down. Exchanges were the prototypical example of a belt-and-suspenders approach to data processing.

To reinforce this reputation, exchanges were one of the first to bunker their data centers and create dual processing facilities. They were among the largest customers of nonstop, fault-tolerant technologies, allowing the exchanges to operate at reliability levels of 99.999 percent, allowing for less than 5 minutes of downtime a year. To achieve this level of reliability, each computer that routed, matched or disseminated orders, trades and market data typically was deployed redundantly, causing each transaction to be executed twice before it was complete so if a problem occurred with one hardware set, the other set would continue without fail.

Today, most -- if not all -- the nonstop, fault-tolerant hardware is gone. The bunkered data centers -- gone. The belt-and-suspenders approach -- gone. And unfortunately, so is our market's reliability, as virtually all of the exchanges over the past few months have experienced significant technology challenges.

What happened? Why the change in philosophy? Where did the belt go? Where did the suspenders go? Why are the exchanges' preverbal pants down around their ankles?

You want to know who to blame? Look in the mirror.

If you haven't noticed, the exchange business has radically changed over the past few years. Exchanges have gone private, costs have been slashed, message volumes have skyrocketed and competition has forced exchanges to rethink their business models.

Bunkered data centers are expensive; fault-tolerant hardware, staff, belt and suspenders -- all expensive. Running a technology organization at 99.999 percent reliability cannot be done on a shoestring budget.

In addition, competition has come to the stayed exchange world. ECNs that were run on a shoestring emerged, and in fact, many prided themselves on their low-cost, high-speed infrastructure, poking fun at the stodgy exchanges with their slower, more deliberate and more expensive architectures.

While cost is certainly an issue, we can't blame cost-cutting alone for the recent poor exchange reliability. Volume and speed are also at play. Rapid message growth and latency have forced exchanges to reexamine their technology architectures, pushing them toward parallelization methodologies to gain speed, lower cost and remain consistently up. Well, two out of three ain't bad, is it?

And, believe it or not, we voted for this. We voted for cheap and fast over reliability. We told exchanges that we preferred ECNs' low-cost / high-speed infrastructures, and if they go down for a little while, well, we can always go back to the exchanges -- they'll be there, right?

Wrong again. The ECNs are now the exchanges. Nasdaq closed its Trumbull data center and consolidated its technology onto the Island ECN infrastructure. The NYSE continues to run the Archipelago ECN while it is rearchitecting its matching engine to automate the floor, increase throughput and reduce cost. And BATS is run out of a strip mall in Kansas City (actually, its technology is hosted in a Jersey City datacenter -- but it certainly doesn't have the resources of either two majors).

So what can we expect? We should expect -- and actually demand -- that our exchange and ECN infrastructures be more robust. If that means buying more hardware, changing designs or replacing IT management, then so be it. But the exchanges deserve our patience, as the exchange business is changing day by day, message rates are doubling year on year and latency is playing a larger role in trading.

Unfortunately, however, patience is a virtue few on the Street have, and while we are waiting for the exchanges to get it right, the ones that do will certainly be rewarded. And anyway, while the exchange may be our symbol of capitalism, competition is really its heart.