Federal Reserve and Mainframes

Over on the Mainframe Executive blog, there is an open letter to the US Federal Reserve Bank, questioning the Fed’s apparent desire to move or switch their systems away from mainframes to distributed systems. Well you would expect less from the Mainframe Executive blog. I have a different take on why the Fed should not only keep their mainframe, but why they might want to move more work to it.

I worked on many of the early mainframe Internet applications. I did the high level design and oversaw the implementation of an Internet Banking Solution that the bank, Sun Microsystems and Microsoft had all failed to get to scale. Our design went from 3k users to I believe at the end of 2-years in production, close to 990k users without an upgrade, and without a system outage. It was built off two mainframe systems outside the firewall, running as a Sysplex. I also did a design review for a bank that had lost close to $60k from four accounts, the back end on the mainframe the mid-tiers and Internet servers distributed.

The point of this post though isn’t to gloat about my success, isn’t being a ‘mainframe bigot’ or even saying the Fed should use the mainframe. In the Mainframe Executive they raise the usual specter of security, yes security is a big deal for banks, even more so for the Fed. So yes, make a big deal of it.

However, the single most important thing to understand about building trusted computing systems, isn’t that you provide a 100% secure environment, in which applications aka business transactions, run. It is that you can show who did what, when, and how. Auditing is much more important than security. If you believe you have a 100% secure system and you lose some money but can’t audit it, what do you do, shrug your shoulders and say “oh well never mind”?

Auditing isn’t about just seeing that you have procedures in place. It is the ability to pick apart a debit transaction on a system that was executed at 4:05pm along with 30,000 others, show how that transaction was invoked, where from, under what security context, what ID, and the originating network address and more. That might require looging through logs of 7-10 distributed systems.

If like the bank I did the design review for, you can’t show the correlation of events leading up to the execution of the transaction, and you don’t know for certain where the user eneterd the network, what ID they used, and how that security context was passed from one system to another, then you don’t have security, no matter what they say.

When you are looking after the nation’s money, and despite the obvious current finicial position of the US, budgets not withstanding, I’d say that was pretty important. What does the Fed say?

I say “Show me the audit, show me the audit, show me the audit…” (repeat ad infinitum)