Cian Kenneth Ross
[Send email using form at bottom]
June 18, 2008
PROFILE
Software system architect, senior software engineer, quality-assurance
engineer, and software master craftsman seeks challenging problems to solve by
architecting and building well-engineered solutions. Dedicated to creating
the right solutions built and documented in the right ways.
Experienced in C/C++, Perl, shell scripting, common UNIX/Linux tools, Linux,
BSD UNIX, SunOS, Solaris, and in writing code that runs on multiple OS and
hardware platforms. Known as the "go-to" person for questions about Perl, C,
UNIX, and Linux.
Thrives on taking projects all the way from initial conception and analysis
through architecting, design, implementation, testing, rollout, and user
training.
KEY ACCOMPLISHMENTS
Architected and built a new scheduler and architected the new active agents
for Cadence Design Systems's internal product gridMatrix. Extracted twice to
three times the previous performance from their compute farms. Wrote
specifications and other internal documentation.
Architected and built a communications layer on top of TCP/IP and Berkeley
Sockets, as a replacement for COPE CORBA in the follow-on system. The new
layer exploited the natural asynchrony of the medium and thus enabled a
higher-performance, fully event-driven style of programming.
Architected and built a system that doubled the throughput of RTL and
gate-level simulations for AMD K7 and K8 parts on the existing compute farms.
The system managed thousands of concurrent simulations per site on Solaris,
Linux, and AIX. Wrote the specifications and the end-user documentation for
the system. Trained system users.
With the same system, reduced the manual-rerun rate for K7 and K8 simulations
by better than an order of magnitude, saving several man-days per tapeout.
Freed a full-time person for other tasks by ending the need for hourly manual
monitoring and intervention.
WAN-enabled the second phase of the system to provide reliable multi-site
execution of simulations and test generators without using NFS mounts or
manual file transfers.
By re-using several of the classes implemented in Perl, reapplied the system,
by writing an additional driver set, to replace an antiquated system for
running tests on actual K7 and K8 parts with a new high-reliability,
low-maintenance solution. Freed another full-time person from continuous
monitoring to do analysis of the results.
PROFESSIONAL EXPERIENCE
Contract at Cadence Design Systems, Inc., Austin, Texas
December 2005 -- Present
Architected and built a high-performance job-scheduling system for the
gridMatrix grid-computing system, to add priority scheduling, fair-share
scheduling, and explicit resource control, tracking, and allocation to
gridMatrix, with multi-cluster support. This included moving gridMatrix
from statically assigned lists of tasks to be run on the execution hosts to
dynamically placing tasks on hosts via a new active agent that runs on each
host and manages activity there. This also included architecting and
implementing error-recovery techniques to deal with a variety of issues that
cause job failure, to include automounter problems, network partitions,
failures in DRM software such as LSF, Condor, and SGE, and "black hole"
hosts that appear to accept work but then do not produce results. The new
components plug in to the existing code via the scheme already in use in
gridMatrix with COPE CORBA.
Architected and built a replacement communications layer after the higher
performance of the new system revealed a variety of things left to be desired
in the performance of COPE. The new layer offers the full capabilities of the
underlying sockets layer, to include full use of the natural asynchrony of
network communication. This then allows a fully event-driven style of
programming, allowing the elimination of certain kinds of delays due to
temporarily blocked processes, and without the use of heavyweight techniques
such as multithreading.
Contract at Motive, Inc., Austin, Texas
June -- September 2005
Made Motive able to rely realistically on a critical build system by porting
the system (about 120,000 lines of Perl code) from Apache 1.3 and mod_perl 1
to Apache 2.0 and mod_perl 2.0.1, both using MySQL 4. This made the company
able to run the system on faster, up-to-date hardware for which maintenance
contracts are available. Reduced scheduling times of large builds with many
dependencies from half an hour or more to less than a minute. Made the system
far more easily maintainable and configurable by moving its configuration data
out of hard-coded constants in the source and into a per-host master
configuration file. Built the object that handles this data. Removed
unnecessary complexity from the code and pruned out old code that had already
been removed from use. Fixed a variety of bugs. Wrote the specification for
a new communications architecture between the scheduler and the official-build
machines with a view to improving reliability and further decreasing the
complexity of the system. Documented the large number of dependencies on
external modules, particularly where specific versions of those modules were
required for correct operation of the system. Documented both the old and new
installation, configuration, and maintenance procedures.
Independent Software Development
2003 -- May 2005
Architected and built an initial mockup of a different sort of deep-space
flight simulation, plus an initial cut of flight-dynamics code, targeted at
the computer-gaming market, with a view to finding the right investor when the
project is further along. This project combines experience with client/server
systems (to support multi-player modes) with climbing new learning curves such
as computer graphics and celestial mechanics. It includes Object Oriented
analysis and design, plus extensive coding in C++ and an XML-like language.
Support code in OO Perl handles parsing of the XML and translations to HTML
pages. Specifics about the project itself are available under non-disclosure.
Advanced Micro Devices, Austin, Texas
2000 -- 2002
Architected and built a new system that doubled throughput for K7 and K8 RTL
and gate simulations on the existing compute-farm hardware. The second phase
of the system automated reliable cross-site data transfer over a WAN.
Architected and built the web- and Oracle-based bug-tracking and
statistics-reporting systems for gate simulations, using Perl CGI scripts.
Maintained and upgraded the parallel system for the RTL simulations.
Ran time-critical RTL, gate, and tester-vector production on multiple Athlon,
Duron, and Hammer parts, usually concurrently, with mid-tens of thousands of
patterns and vectors per version of a part, requiring as long as three weeks
to run full K7 regression sets.
BMC Software, Austin, Texas
1999 -- 2000
Tested the server side of LogMaster. Analyzed specifications and built a
regression-test suite. Built a test framework that supported parsing of
the SQL output, allowing the tests to verify it without using literal-text
captures. Found more than 200 issues prior to release.
Architected and built a distributed system that automated LogMaster testing on
multiple versions of AIX, HP-UX, and Solaris, and multiple Oracle versions.
Ran more than 2200 combinations of test, platform, and Oracle version in a
24-hour turnaround time, largely eliminating manual regression testing. Built
a web/CGI interface that gave casual users an easy way to access and manage
the system.
Architected and built an automatic test generator that created tests and test
data for LogMaster. Found more than 100 issues in addition to the results of
the hand-written regression tests.
Tested LogMaster installation procedures and end-user documentation.
Texcel Ventures, Austin, Texas
1997 -- 1999
Technical lead and architect of the solution for server-side QA of IM, a large
OO database application for SGML and XML document management. Set Texcel's
direction for server QA. Specified phased hardware and software requirements.
Handled functionality, regression, and stress testing. Benchmarked IM.
Worked closely with the developers through all phases of the product cycle.
Automated the running of regression tests, stress tests, and benchmarks,
across multiple platforms.
Wrote IM server test plans and tests. Analyzed the C++ API, including hundreds
of methods and dozens of object classes, on an ongoing basis to ensure
continuing test coverage and best use of testing resources. Wrote most of the
tests, mostly unassisted due to personnel constraints. Supervised the writing
of the remainder of the tests. Ensured that the C++ test code remained
platform-independent. Found more than 300 issues prior to release.
Architected and built Texcel's bug-reporting and -tracking system and
feature-request system based on IM plus web/CGI forms. Built the SGML DTD for
storing this information.
Built core-application and API code for managing SGML entities, including
versioning and selection.
Contract and Freelance Projects, Austin, Texas
1991 -- 1997
Tandem Computer Corporation
February -- October, 1997
Tested the DGM&S OMNI SS7 on Puma, Helix, and Dual Node Puma platforms, plus
related OS and hardware functionality. Wrote test applications, scripts, and
support files.
Advanced Micro Devices (Luther Team)
1995 -- 1996
Architected, built, and ran a system for running high-volume K5 RTL
simulations on 500 HP/UX workstations, for an order-of-magnitude throughput
improvement. Achieved half a billion simulation cycles per month. Eliminated
hourly manual monitoring of production runs. Did the initial analysis and
qualification of the test results.
Advanced Micro Devices (Directed-test Group)
May -- August, 1995
Built directed tests in assembly language for 80x86 microprocessor
instruction-set verification. Analyzed specific instruction-set features and
wrote test plans.
Wayne-Dresser
January -- April, 1995
Updated internal specifications and program documentation for an embedded
system from code written in 8051 assembler, Z80 assembler, 80x86 assembler,
and C.
Tandem Computer Corporation
June -- August, 1994
Documented Tandem's port of the gated implementation of the Open Shortest Path
First (OSPF) protocol, with heavy reliance on the gated code for information.
Tivoli Systems Inc.
1992 -- 1993
Wrote API documentation and wrote and documented new customization-code
examples for early TME versions, with FrameMaker. Relied heavily the source
code for information.
DTM Corporation
1991
Maintained and built critical software components for equipment that
implemented the company's proprietary process.
University of Texas, Austin, Texas
1980 -- 1990
Built OS-level software for a custom-written system on CDC mainframes,
including an order-of-magnitude performance improvement for a system utility,
and both the FTP client and the FTP server. Wrote custom UNIX applications,
including a replacement for an obsolete RJE system (this project included BSD
kernel code), and other network-based projects.
EDUCATION
Bachelor of Arts, major in Computer Science, University of Texas at Austin
PERSONAL ACCOMPLISHMENT
US Private Pilot certificate, with instrument-airplane rating
|
