Syniverse Technologies (2001–2016)

Syniverse was the backbone infrastructure provider for the global cellular industry — handling roaming, number portability, billing clearinghouses, and real-time messaging data for virtually every major carrier. Fifteen years here meant working at massive scale, with real-time systems that could not go down.

Sr. Systems Engineer (2001–2010)

My first decade at Syniverse covered a broad range of telecom products: STARS (Streamliner Terminal Access Reporting System), CNAM-LIDB (Calling Name & Line Information Database), MDR (Mobile Data Roaming), AAA Broker, and the VisWise/PanWise/UniWise data visualization suite.

In-Memory Carrier Lookup Server (C++)

One of the more satisfying problems I solved in this era was replacing a slow Oracle-backed lookup function in the MDR system. The input was a network node ID; the return value was the carrier record for the company handling that session.

The original implementation made a synchronous SQL query over TCP — expensive not because of network distance (everything was in the datacenter) but because Oracle is Oracle, and round-trips add up at scale.

The replacement was an IPC-based server process that held the entire dataset in memory using mmap(). The data file was millions of binary, fixed-length records, pre-sorted on the key field — so rather than parsing anything, the server called bsearch() directly on the memory-mapped region. Zero heap allocation, zero parsing, O(log n) lookup.

Hot reloading was handled by a background thread that polled the file's inode timestamp every 60 seconds. When the file changed: a new mmap() was established, in-flight threads picked up the new pointer on their next iteration, and the old mapping was released with munmap() 60 seconds later. Zero-downtime data refresh.

The result: lookup latency dropped from milliseconds to microseconds.

High-Volume Oracle Data Loaders (C++)

For the VisWise product family, I wrote most of the high-volume Oracle data loaders. A recurring theme in this work was finding where the system was fighting the database unnecessarily — row-by-row inserts, suboptimal query plans, missed bulk-load APIs — and replacing them with approaches that worked with the database engine rather than against it.

DHCP Load Testing Tool (C)

For a large in-carrier DHCP deployment within MDR, I wrote a load testing tool in plain C that simulated thousands of simultaneous DHCP sessions within a single process. Sometimes the right tool is still C.

Tandem/Guardian Middleware (C)

For the Streamliner system, I wrote middleware for a Tandem/Guardian mainframe that enabled network access to SQL/MP (the Tandem database). This was one of those projects where the interesting challenge wasn't the algorithm — it was understanding an entirely different computing paradigm and bridging it to the rest of the stack.

Technical Lead, VisProactive (VPA) (~2010–2014)

VisProactive was Syniverse's flagship real-time analytics platform for cellular carrier networks — ingesting and visualizing billions of rows of signaling and messaging data per day.

60x Data Loader Performance Improvement

The existing data loader was a shell script invoking Vertica's basic CLI tool once per file. Every invocation paid the full cost of process startup, TCP connection, authentication, and session setup — for every file.

I rewrote it as a persistent Java process using Vertica's streaming bulk loader API directly. Eliminating per-file setup overhead alone yielded a 20x throughput improvement. Running three independent loader threads — each pulling from a shared work queue with deduplication to ensure no file was processed twice — tripled it further.

The resulting system processed over one billion rows per day, with headroom estimated beyond two billion.

Resilient FTP/SFTP Distribution Client

VisProactive distributed terabytes of cellular messaging data in real-time, 24x7, via a multithreaded Java FTP/SFTP client. I added comprehensive failover logic covering: connection timeouts, connection refused errors, interrupted transfers, and out-of-disk conditions on target servers (detectable via write errors).

Rather than hardcoding thresholds, I implemented the failover conditions as a Rhino (server-side JavaScript) script embedded in the JVM. When operations staff wanted to tune thresholds — and they did, constantly — they edited the script themselves. No release cycle required. This was an explicit decision to get out of the business of trivial operational changes.

(I did also advocate strongly for replacing unauthenticated FTP with SFTP. The response was that it was all internal networking, so "no one will get it." Famous last words.)

Frontend Modernization

Led the proof-of-concept work that drove a company-wide migration from legacy Adobe Flex and Oracle ADF applications to AngularJS. Also built JobMonitorWebServer, a custom jQuery application that let operators start/stop processing threads, change log levels, and tail log files in real-time — all from the browser.

Platform & Infrastructure Work

Technical Lead, Consumer Insights (SCI) (~2014–2016)

SCI was a new business unit enabling non-telecom clients (banks, retailers) to reach their customers via mobile handsets using anonymized behavioral propensity models — delivering targeted offers when customers were physically present in relevant locations.

Key contributions:

← Back to Work