The evaluation ran through the winter of 2006 and 2007 against the incumbent network manufacturer: one side positioning a full application proxy with universal inspection, traffic profiles and a programmable rule language, the other a load balancer.
The question underneath was whether the City wanted a device that balanced traffic or a device that could understand and rewrite it.
A written question list went to the manufacturer rather than working from the pitch: which layers it functions at, whether it covers all seven, how access logging is configured and what enables it. The logging answer — standards-compliant access logging through the acceleration module, or request logging written as rules — turned out to name the mechanism that carried the next decade of work on that platform.
An evaluation is worth running only if the answers change the design. Those answers did.
Specifying the application security requirement at design time
With the order placed, the question asked was whether the platform could front internal web services — proxying a web services call through to the application server behind it — and whether an application module was needed to inspect what was passing through.
That is the injection problem stated as an architecture requirement rather than discovered later as a security finding. Putting a reverse proxy in front of a web service helps only if something in the path is actually reading what crosses it.
The application security module went onto the purchase order alongside the load balancer because of that question, instead of arriving as a separate security purchase years later. The record supports it as specified and purchased; the last word in the record is an integrator noting the module still had to be built, so what is claimed here is the requirement and the procurement rather than a configured control.
What it has to do gets settled before where it goes
No installation date was set until a design session with the manufacturer’s own engineer had happened, and the unresolved questions were named rather than waived through — including whether the platform supported the intended topology, and what the default gateway arrangement would be.
The default gateway question determines how every balanced server routes. That is not an install-day decision, and a deployment scheduled before it is answered is a deployment that will be answered by whoever is holding the console.
Programming it
The platform retired an appliance-based reverse proxy and web access product, and its implementation went into the infrastructure group’s annual accomplishments while that group was managed from this desk.
The rules were written here. A host match redirecting to secure transport for the building inspections site in 2007, and the harder version of the same pattern worked through a manufacturer support case. The configuration backups were kept here too, certificates and stream configuration included.
Over the following years that rule set carried analytics logging, source-address redirects, the public wireless landing page, event site cart redirects matched to the certificate configuration, and access control.
That is programmable infrastructure a decade before the phrase was current in that organization: a migration could be cut over or a site retired by changing a rule on a traffic manager rather than by touching an application.
It also concentrates risk in one device, which is the honest other half. When redirect rules were disabled in 2014, the question taken to the security information platform was what was still arriving at the old addresses — because a rule that has been quietly carrying traffic for years is load-bearing whether or not anyone remembers writing it.
Configuration that lives in data rather than in logic
The public wireless redirect was designed so that adding a building was a list entry rather than a change to a production rule. The virtual server does not know which buildings offer public wireless, and neither does the rule: a data group holds the address ranges and the rule consults it. Extending the service to the courthouse in 2013 meant adding an entry.
The closure note on that request recorded that the range was in fact already partly covered by an existing mask — which is exactly what a shared list makes visible and what scattered per-site rules would have hidden.
Configuration that lives in data rather than in logic is the difference between an operator adding a site and an engineer editing production.
Sixteen years, ending as continuity infrastructure
The production pair was refreshed and relicensed, taken through a major version upgrade in 2022, and carried in five-year planning toward virtual appliances on subscription terms.
On the disaster recovery five-year plan the traffic management layer is the mechanism that answers the continuity question: global traffic management deciding which data center a request reaches.
Bought as a server proxy in 2006 and ending as continuity infrastructure, which is the usual life of a platform chosen for what it could be programmed to do rather than for what it did on the day.
Outcome
Programmable infrastructure a decade before the phrase was current in that organization — and the mechanism the continuity design eventually rested on.