Tom Keiser Email & Phone Number

San Jose, California, Us

San Jose, California, Us

Edgewise was acquired by Zscaler in May, 2020. I am presently the software architect for the Edgewise product.

Burlington, Massachusetts, Us

I view my role as supporting our engineering organization in productization of big ideas. At the 50,000 foot view, the topology of a product organization such as ours can be distilled—by ignoring feedback loops, etc.—to its essence: business needs/pain points → brainstorming → prioritization → innovation → engineering → QA → customer.If one zooms in on the innovation to engineering transition, risk parameters are at least as important as prototypes. Modern SDLC processes tend to thrive under low variability. Much like Taylorism and 6-sigma, scrum-like processes rest upon accurate prediction of work unit duration in order to maintain consistent throughput. In a greenfield, however, predictability is illusory: considering project time-to-completion in statistical terms, the mathematical expectation value of duration differs markedly from human predictions—because the underlying distribution is a power-law! In my experience, this impedance mismatch is not merely vexing to management; it undermines team confidence as it thwarts forward momentum.Many counter that agile research spikes are the solution. Such techniques work well in engineering organizations implementing well-understood, widely-solved problems. However, applying the proverbial hammer to greenfield problems is often inefficient, and disillusionary—as spikes aren't suited to systemic risk exploration. Rather, the innovation organization must de-risk large projects by seeking out and clipping the right tail risk of a project's duration before engineering embarks on its journey.This is easier said than done. For when I say 'risk', I am technically conflating risk with uncertainty. Both apply here. The key is to brainstorm in order to reduce unknown unknowns and imagine failure modes, then design/run risk mitigation experiments. The end result: not mere prototype(s), but crucially a "topographical map" of the problem space—informing engineering's planning process by illuminating pitfalls.

Burlington, Massachusetts, Us

Edgewise Networks provides Trusted Application Networking for the cloud and data center to protect where firewalls fail. Edgewise prevents progression of network-borne threats by allowing only trusted applications to communicate over approved network paths. Using machine learning, Edgewise recommends and prioritizes optimal policies to eliminate 98% of the network attack surface and protect the rest.

OneCloud was a VMware to AWS disaster recovery startup. As the second hire, my principal responsibility was to serve as a storage architect. I designed and wrote our data storage and asynchronous replication layers -- with the key goal that all the components eventually come together as a asynchronously-replicated, cloud-cost-optimized, deduplicated, fault-tolerant, geographically-distributed file system, specifically optimized for storing large objects (VM disks) whose changes needed to be tracked in granular increments (64KiB chunks on VMFS) for space and transmission efficiency reasons. A significant fraction of my time was devoted to the optimization problem of minimizing aggregate cost across EC2+EBS, S3, and EBS Snapshots in an environment where EC2 and EBS were spun up on demand (to process recently uploaded incremental backups).Additionally, I: 1) led our efforts around VMware VADP related development and integration; 2) was responsible for packaging our FreeBSD-based appliance kernel and userland stack; 3) performed kernel and ZFS development as required; 4) was heavily involved in our devops automation efforts; 5) led many corporate IT initiatives and authored most equipment procurement proposals; 6) tried to lead by example in terms of developing rational system/software architectures that intrinsically scale; and 7) served as a system, storage, and network admin for most of our on-premise and colocated infrastructure.

San Jose, California, Us

I was a member of the Nexenta platform engineering team, which develops improvements to the Illumos kernel. The majority of my time was dedicated to advising the exec team on the technical feasibility of a couple new initiatives piloted by the CTO's office, as well as advising on technical due diligence for a potential business transaction. Similarly, I assisted with technical due diligence of one of our OEMs.When writing code, I worked on the kernel CIFS server, HA clustering, ZFS, the SCSI target framework (COMSTAR), and the Kerberos client. Additionally, I served as a technical spokesperson during a prolonged debugging effort with a marquee customer, which had escalated up to the customer's exec team, requiring assiduous and precise communication.

Leesburg, Va, Us

Leesburg, Va, Us

During my tenure as an AFS engineer, I worked on two major initiatives: a large-scale rewrite of the OpenAFS file server, and a dynamic distributed systems instrumentation framework for OpenAFS. The primary motivation for the file server rewrite was a series of operational problems resulting from several-hour file server restart times. My modifications (please see my conference presentation[1] for further details) resulted in a server recycle time that was three to four orders of magnitude less than the legacy file server, yielding substantial improvements in plant stability for large AFS deployments. In addition, these modifications improved AFS-3 cache coherence protocol overhead by no longer resetting client state across server restarts.Secondly, I was the principal developer for a large initiative to build a distributed systems dynamic instrumentation framework for OpenAFS (please see my conference presentation[2] for further details). Although the changes proved too daunting for acceptance by OpenAFS, the core ideas derived from this project yielded a sound theoretical basis for building scalable distributed systems dynamic instrumentation frameworks (by applying speculation, asynchrony, and aggregation), which I later expounded in an impromptu presentation given at the 2008 DTrace conference.[1] http://workshop.openafs.org/afsbpw06/talks/tkeiser-dafs.pdf[2] http://workshop.openafs.org/afsbpw07/talks/tkeiser.pdf

Armonk, New York, Ny, Us

I was involved in a collaborative research project between IBM Research and the Linux Technology Center. Our goal was to develop an autonomic test scheduling framework for the LTC's kernel testing datacenters. A number of interesting concepts came out of this effort, such as utilizing the Lcov framework to develop parametric, statistical line coverage fingerprints for each test case within the LTP suite (indexed by each supported hardware architecture, and a few other relevant parameters). These fingerprints then went through some simple signal processing in an attempt to remove common-mode noise, e.g., timer interrupts unrelated to the test case. This post-processed coverage data could then be matched against changed lines in a diff in order to assign priority weightings to test cases, thus--at least in theory--tightening the feedback loop for regression testing. Additionally, we were working on ways to leverage genetic algorithms to explore parametric state spaces within more complex unit tests, and then attempting to prioritize exploration of specific regions of state space based upon historical data.

Baltimore, Maryland, Us

This was a high school internship position during which I was tasked with designing a database to track all AC electrical loads within several of PECO's (now Exelon's) nuclear facilities. In addition, I designed a series of visualizations to aid engineers and control room staff in estimating breaker, bus, and step-down transformer loading under numerous feeder breaker positions, plant operating modes, and design-basis accident (DBA) scenarios.