(last updated - 2026/08/14)
                                Dennis Towne
                       11321 South Grandville Ave #301
                          South Jordan, UT. 84009
                             Tel: 716-462-8270
                            email: soda@xirr.com

                              Work Experience

   Aug 15 - Current; Google, Inc.
   General
    - Primary responder for the company-wide emergency response team
      'Tech IRT'; responder on ~40 major incidents over six years,
      including datacenter power, cooling, and hardware failures, large
      network outages, and large-scale security incidents, many of them
      affecting ML fleets and capacity
    - Site dean for SRE EDU onboarding education; taught ~150 classes
      to roughly 1500 new engineers, and as dean ran quarterly
      week-long onboarding sessions end to end for the Pittsburgh
      office

   Site Reliability Engineer - 2021-Present - Google DeepMind / ML Serving
    - Five years of backend oncall for Google's production ML serving
      stack: Bard, Gemini, Veo, and nearly every other major model
      launch, on a fleet of tens of thousands of model server jobs
      serving thousands of models for Search, Ads, YouTube, and other
      product areas
    - Designed large parts of the load balancing and the entirety of
      the load shedding configuration for the initial Bard LLM
      launches; since then, owner of autoscaling design and
      configuration for model servers and routers across the fleet
    - Built load shedding, autoscaling, and load balancing as one
      integrated system: shared overload signals (CPU/TPU utilization,
      backlog, contention estimates) drive both shedding and scaling,
      load balancing spills traffic away from unhealthy tasks without
      starving the autoscaler, and default configurations cover
      thousands of heterogeneous models with little per-model tuning
    - Design goal: no task ever falls over, regardless of offered
      load.  Shedding is designed to survive ~10x overload; in one
      major incident the fleet absorbed a sustained 3x overload for
      several hours at 100% accelerator utilization with only slightly
      elevated latency
    - Deep familiarity with accelerator serving pathologies:
      non-interruptible one-request-at-a-time TPU pipelines, CPU-side
      request batching, keeping pipelines packed without backlogging,
      and isolating high-latency from low-latency traffic

   Site Reliability Engineer - 2017-2021 - Ads Trust and Safety
    - Team tech lead; responsible for the SRE engagement with the
      browser spam detection systems
    - Owned reliability for four large log joining and log processing
      pipelines that ran a battery of filters against browser traffic
      logs to flag spam clicks and impressions.  Each pipeline ran in
      duplicate in two geographic clusters, and a separate layer built
      on a large distributed Paxos cluster deduplicated and merged the
      output
    - Pipelines had a target processing latency of under an hour, but
      could fall hours or days behind during incidents and then have
      to recover quickly.  Much of the work was capacity engineering
      for recovery: resource planning ahead of demand, scaling work to
      keep the systems moving when backlogs became unmanageable, and
      a migration to SSD backing storage that bought roughly 2-3 years
      of additional lifetime while the replacement was built
    - Wrote the templates and configuration for the replacement
      pipeline, built on more modern log joining infrastructure, that
      consolidated the filter pipelines and the deduplication layer
      into a single system.  Once it was in place the SRE engagement
      was no longer needed and was wound down
    - SRE for the ads integrity system, a large handler-driven
      platform that checks and ultimately approves or rejects ad text
      and ad content submissions that violate the Google Ads terms of
      service

   Site Reliability Engineer - 2015-2017 - Google Play
    - Help keep Android Play running
    - SRE Service owner for Play Games
    - C++ and Java backend maintenance work
    - Build tooling and infrastructure for Android and ML backend services

   Apr 03 - Apr 15; Vocal Technologies
   Associate - electrical engineer/ lead programmer
   Languages - C, various DSP assembly languages
   Duties/experience:
    - Linux development - ARM/MIPS/x86 device drivers, application
      layer, boot scripts, debugging, etc.
    - SIP protocol stack implementation
    - Cryptography - create specification for patient unique identifiers
      for Maryland State Health Care Commission (MHCC)
    - Design and implement AES/CCM cryptography subsystem as Linux
      device driver for ad-hoc mesh network packet radio
    - Implement MIL STD 16/32 kbps CVSD vocoder from specification
    - Develop signal processing library for medical artery elasticity
      measurement device
    - TI and ADI Blackfin DSP hardware bringup
    - ADI, TI, ARM, Microchip, and other DSP programming
    - Subband-based AEC (Acoustic Echo Cancellation) development
    - G168 echo cancellation algorithm development
    - Implement software encoder/decoder for (16x11)^2 turbo product code
      using extended Hamming subcodes
    - Data modulation maintenance and porting, V.21, V.23, V.22,
      V.32, V.34, V.90 and related data protocol schemes
    - Fax protocol and pump maintenance, V.27, V.29, V.17, V.33,
      V.34HD, T.30, T.31, T.32, T.4, T.6, and more
    - ILBC floating point to fixed point conversion
    - Voice coder performance optimization
    - PSTN FXO/FXS hardware design and bringup
    - Narrow band FM/FSK modulation/demodulation for RF links
    - RF hardware design (400 and 800 MHz range)
    - ATA hardware and software development

   Jan 95 - Current; Dentinmud Internet Services
   Owner - privately held company
   Languages - C/C++
   Duties/experience:
    - Design, create and maintain multi-server MMO codebase in C/C++
    - Develop and maintain cross platform libraries for Win32,
      MacOSX, and Linux
    - Creation and maintenance of cross platform GUI client
    - Responsible for web site, content additions, online servers
    - Manage backups, upgrades, and security fixes on servers
    - Design new additions and content
    - Improve accessibility for visually impaired users
    - Manage staff of ~20 volunteer builders

   Jun 97 - Sep 02; IBM, eServer division
   Software engineer for iSeries OS/400 operating system.
   Languages - C/C++, PPC assembly, PL/MP
   Duties/experience:
    - Design, add, and debug core operating system routines including
      interrupt handlers, process management, synchronization
      primitives, and low level statistics collection
    - Implementation of hardware multithreading support
    - Use DEV/2000 RCS as part of a strict defect process
    - Performance and systems analysis of OS/400 workloads

   Apr 96 - Apr 97; Geography Department at Utah State University
   Lead programmer for Mojave Georeferenced Bibliography Database.
   Languages - C/C++
   Duties/experience:
    - Design and build a georeferenced database for research data, along
      with parsers for adding new entries and an HTML interface for access
      to the database. The majority of this project was written in GNU
      C++, and was ported to a number of operating systems, including AIX,
      ULTRIX, Linux, SunOs/Solaris, and Windows 95.

   May 94 - Sep 95; Utilities Department at University of Alaska Fairbanks
   Full time temporary assistant to site electrical engineer
   Duties/Experience:
    - Experience with Allen-Bradley Programmable Logic Controllers, wiring
      and installation of Square D 'PowerLogic' networking system. Was
      responsible for setup and programming of Square D circuit monitors
      and maintaining/upgrading network. Extensive work on General Electric
      switchgear and 10 MW steam turbine. Troubleshooting of ground faults,
      static excitation system, and control/metering circuits.

   Dec 93 - Jan 94; Professor Leroy Hulsey
   Assembly language programmer
   Languages - x86 Assembly
   Duties/Experience:
    - Write, test and debug assembly language library for use with Lahey
      Fortran. Routines included interfacing with HardLock security devices
      attached to printer ports, interrupt handling, disk I/O, writing
      mouse routines, and direct screen buffer access. Was responsible for
      final packaging of a product, and security on said product.  Help
      facilitate upgrade of software to run under Windows 95.

                            Miscellaneous Skills

   Large scale distributed systems: load balancing, load shedding,
   autoscaling, and overload protection for ML serving fleets
   ML/LLM inference serving on TPU accelerator hardware
   Production incident response at datacenter and fleet scale
   Extensive experience with C and C++
   Experience with other languages, including a number of assembly
   languages and scripting languages such as Python, Perl, PHP,
   and Bash
   Algorithm optimization
   Signal processing, including telephony, voice, and data modulations
   Cryptographic system design and implementation
   Hardware bringup of new custom boards
   RF hardware bringup/troubleshooting/analysis
   Algorithm implementation using fixed point math
   Creation of cross platform, end user GUI applications using WxWidgets
   Unix/Linux system administration 
   Unix network and systems programming

                                 Education

   Spring 1997; Utah State University
      BS in electrical engineering, computer emphasis.  Minors in math,
      physics, and computer science.