Drew Eckhardt Email & Phone Number

Mountain View, Ca, Us

Mountain View, Ca, Us

System Software, Distributed Systems: Private cloud storage.Uber technical lead handlingGoogle Distributed Cloud connected Data Center storageGDC air-gapped HA + Multi-Zone Disaster RecoveryGDC storage control and data plane re-architecture

Redwood City, Ca, Us

System Software, Distributed Systems: SaaS Cloud Content Management. Making Box enterprise grade at $1B in ARR.Box Canvas virtual whiteboard architect. Supported off-shore team integrating with the Box platform. Led distributed real-time gateway project routing client-to-client, client-to-server, and server to client messages over WebSockets.Led Microsoft Office on Box storage coauthoring data path project and delivered on an aggressive schedule. Leveraged optimistic concurrency control. Streamed content with parallel read ahead and write behind.Conceived and drove Policy Driven Observability program across a 500 engineer organization detecting 80% of critical feature site events, triaging 60%, and improving MTTx. Service owners specify signals, thresholds, and infrastructure used; infrastructure engineers provide standard signal packages; SREs implement policy transforming that to dashboards and alerting following best practices; change code review by service team + SRE insures accuracy; and post-mortem drives improvement.

System Software, Distributed Systems: Scaling whole fleet telemetry handling supporting $2.8B/quarter operations with 1.5 billion daily active usersInitiated and led time series telemetry projects to reduce service capital expense 40%. Modified Beringei in-memory time series database to store 85% of data in flash with hot/cold classification. Reduced operational load with better dashboard.Handled capacity planning and hardware allocation.

Eden Prairie, Minnesota, Us

System Software, Distributed Systems: SaaS enterprise cloud backup product with multi-platform multi-threaded client code, web back-end, and supporting libraries. Post merger integration.Cloud direct backup data path leadArcserve acquired Zetta on June 13th, 2017Responded to market demand for increased backup frequency (reduced Recovery Point Objective) with changes to integrate Arcserve change tracking driver.Designed new parallel hashing scheme to improve RPO for customers with large files combining Merkle trees and Log-Structured Merge trees for persistence with negligible amortized IO load.

Sunnyvale, Ca, Us

System software, Distributed Systems: SaaS enterprise cloud backup product with multi-platform multi-threaded client code, web back-end, and supporting libraries. Pivot from 0 to cash-flow positive 8 figure ARR and acquisition.Data path lead/manager. Moved up market to six-figure ACV customers by making performance Zetta's key market differentiator. Achieved throughput up to 107X faster than competitors. Reduced two day syncs to two hours with client side meta-data caching using only sequential IO and patented the technology. Optimized parallelism and pipelining for maximum throughput with flow control to limit memory footprint.Entered image backup market which became 10% of billable customer footprint. Allowed overnight re-sync of 5TB servers and 200TB data centers. Made Zetta backup to cloud faster than Barracuda to local appliance. Limited COGS with sparse storage. Designed and implemented sub-tree restore within Windows NTFS file system images. Added VMware host side backup to monetize larger Managed Service Providers. Stored images as raw data with translation on backup/restore to leverage our ecosystem including Disaster Recovery as a Service. Implemented change-tracking driven syncs with write aggregation to maximize overnight backup window capacity.Made cloud business continuity viable with low-down time incremental switch-back to customer site.Minimized new feature time to market with agile process. Delivered minimum sellable products for early traction then optimized based on customer experience. Usually shipped the 1st or 2nd release candidate and avoided regressions through Test-Driven Development including automated tests for every feature and reproducible bug. Achieved enterprise-grade reliability. At one point our biggest bug was in "most if not all" x86 C++ compilers. Wrote HTTP and libc fault injectors. Implemented end-to-end integrity checks to detect latent failures. Improved fault tolerance.

System Software, Distributed Systems: scale-out strongly and eventually consistent object storage on flash. Slide deck to product.Distributed systems lead with local and off-shore developers.Designed and implemented metadata object store and lease manager with distributed eventually consistent (to match a competitor's 2-node product), distributed strongly consistent (via Paxos), and single super node (for simpler testing) modes of operation.Designed and implemented primary copy key-value replication with a recovery process that restarts and checkpoints. Designed test platform and implemented key pieces which allow cluster simulation in a single thread with a synthetic time base and reproducible pseudo-random timing with fault injection. Used model checking to catch problems without specific tests.Designed and implemented 'C' event infrastructure with user specified binding of closures to thread environments to leverage multiple cores and support mixed programming models such as message based concurrency control in the complex cluster management plus metadata path and thread pools with conventional locking for parallelism where workload partitioning was not practical.Designed logging infrastructure with hierarchical control (for instance, all allocations or specific type allocations in shared memory) of output and binary storage to allow customer system trace-level log accumulation in shared memory with flush on local or remote error detection and led implementation. The API appears to developers like printf/syslog, but runs at 300,000,000 log entries per second on an 8 core Nehalem system thus allowing trace level logging at customer sites with negligible performance impact when hundreds of thousands of IOs run each second.Significantly improved system performance. Increased throughput from 429,000 to 548,000 IOPS per node via custom memory allocator. Improved thread scheduler context switch performance by 30%.

Seattle, Wa, Us

System Software, Distributed Systems: eventually consistent object storage and caching with cloud scale.Designed and implemented automated scaling for the Dynamo eventually consistent key-value store implementing Amazon's shopping cart.Worked on a distributed cache combining membership, consistent hashing, and memcached client.Fixed S3 bugs.

System Software, Distributed Systems: scale-out block storage with an iSCSI interface. Scaling with thousands of customers.Eliminated customer iSCSI timeouts on snapshot deletion and introduced meta-data redundancy by designing and implementing a write-optimized crash-consistent transactional circular log with throughput up to 10,000X the existing code with validation via simulation and model checking. Productized with online upgrade/downgrade.Improved read load balancing for a 40% performance increase.Designed and implemented read-ahead for better sequential performance.Improved platform verification and the ability to estimate performance impact of changes in allocation policy, I/O scheduling, and meta-data with minimal investment of engineering resources by writing a simulation tool which generates the same local disk workload a storage server would for a specified user workload.Improved the simulation environment to cover disk failure modes, finding the root cause for a five-year-old data corruption problem and other less severe issues.

Us

System Software.Improved throughput of WiFi bandwidth shaping code for allocating bandwidth between classes of users (free WiFi access, paying customer, employee, etc.).

System Software.Consultant, Contractor.Solved customer software problems - wireless bandwidth shaping, cable set top box memory footprint, DVD output of police dash camera footage

Us

System Software, Distributed Systems: scale-out block storage with iSCSI interface. Slide deck to product.Technical lead for twelve engineer storage server group.Built scale-out block storage appliance targeting Microsoft Exchange deployments with short mean time to recovery, minimal interval between backups, and lower operational cost. Pre-recovered writable snapshots allowed "instant" restore, snapshots were 10ms transactions, and performance matched direct attached disk for the customer workload. Architected system incorporating recovery-oriented computing concepts, reallocate on write snapshots, synchronous replication, and asynchronous replication of selected snapshots.Designed and implemented significant product subsystems totaling over 50,000 lines of code including replication, recovery logic, log structured disk software, pageable B+ trees, event driven infrastructure libraries, and application keep-alive mechanism.Helped grow the engineering staff from just myself to over a dozen engineers through hiring, bringing new engineers up to speed, and mentoring junior people.Drove white-box testing effort and leveraged Extreme Programming agile techniques.

System software: Digital video appliances for broadcast television. Pivot to acquisition.Led networking initiative for the AirSPACE disk based digital video recorder for the broadcast television news market through design, implementation, test, delivery, and customer support phases. This allowed streaming breaking stories during ingest, import/export from Avid non-linear editing systems, media movement between AirSPACE units, and an easy-to-use API for third party systems.Achieved high performance with user space control plane and hardware data plane allowing zero copy operation between gigabit Ethernet hardware and RAID buffer memory. Designed and implemented the asynchronous FTP server supporting multiple connections, pipelined network and disk access, and an object oriented stream description model that made the system easy to adapt to industry standard data formats. The data stream included hierarchal meta-data and allowed for arbitrary audio/video interleaving to match disk layout.Fixed real-time problems within the FreeBSD kernel.

San Diego, Ca, Us

System Software, Distributed systems: highly available cellular base station call tracking & billing interface appliance. Prototype through hand-off to off-shore sustaining engineering.Designed, implemented, delivered to customers, and supported key components including high availability switch over mechanism, persistent clustered calls in progress store, clustered completed call storage, white box testing/developer friendly inter-subsystem interface, hardware error tolerant code, and the underlying generic event driven class libraries.Improved the software process through the inclusion of diagrams and other non-comment documentation in the source tree and automated regression tests.Mentored new graduates.

Boulder, Colorado, Us