Chethan Kumar Email & Phone Number

San Jose, CA, US

Menlo Park, CA, US

Design, build and manage efficient IT infrastructure for Meta's backend services (social graph, ML Inference). Model service growth, procure and manage compute capacity, improve software/system efficiency, develop infra insights. Mentor, partner and a team player.

San Jose, California, US

• Data De-duplication:
• Identified several bottlenecks in de-duplication code, exposed negative performance impact courtesy of specific design choices, worked with dev to fix most of them.
• Negative impact: Changes in SSD behavior (12%), additional metadata lookup (30%), cache contention between data and metadata (57%), memory leaks causing periodic cache prune. Cost-Benefit analysis of cache space.
• Analyzed Thread-specific stack to find the weakest link, effect of processor power-performance optimizations on thread performance using linux “perf” tool.
• Developed analytical models to predict impact of specific space savings features on workload performance and their CPU consumption on different processor platforms (90% accuracy).
• Identified bottlenecks in the code based on call-stack data from perf tool and flame graphs.Increase permissible metadata footprint for De-duplication (Consumers: Dev, PM):

1. Developed algorithms for various features (recommendation engine, working set estimation, and resource sizing) of a storage analytics product (called Architect).2. Designed and developed an automated performance measurement frameworks that would execute different benchmarks - storage (FIO, IOMeter), CPU (kernel compile), Systems (Exchange JetStress.

1. Performance characterization of various software stacks powering PernixData's FVP product including core (I/O paths, multi-threaded execution environment, NFS and VMFS I/O acceleration, I/O flushing to primary storage), adjacent features such as network and memory compression, ordered destaging, adaptive destaging and I/O coalescing2. Developed.

Palo Alto, CA, US

1. Diagnosed obscure performance problems related to memory management, CPU scheduling, OS timing, and networking of monster virtual machines (64 CPUs, 1TB of memory). Identified optimizations to software stack that improved application performance significantly (10+ %)2. Designed and developed a performance measurement framework that deployed a 10TB OLTP.

Palo Alto, CA, US

1. Developed a parser to summarize callstack data from ESX kernel and compute time spent in various kernel functions using Python. Data was summarized based on user preference - aggregate, per CPU, range of CPUs, sorted by functions, sample counts, and grouped by various kernel modules2. Showcased VMware's supremacy in supporting extreme I/O demands for.

Round Rock, Texas, US

1. Technical lead for designing and testing Dell’s pre-validated Oracle database and Real Application Cluster solutions. These solutions included Dell PowerEdge servers, Dell PowerVault storage, Dell|EMC Fibre Channel storage, Windows and Linux operating systems, various RAID controllers, Host Bus Adapters, and Fiber Channel and Ethernet switches.2..

Round Rock, Texas, US

1. Designed and prototyped an automated build infrastructure using “CruiseControl” [an OpenSource platform] to build OpenManage Server Administrator [OMSA] software suite. 2. Wrote scripts for automating build process and various ClearCase operations (make and recommend baselines, rebase, get latest baselines) using NAnt, Perl, and DOS. 3. Debugged and.

Bangalore, Karnataka, IN

1. Led a team of 7 engineers to manage the IT infrastructure of one Wipro's software development centers.2. Worked in a team on an award winning six-sigma project to reduce the defects and improve the process of managing employees’ domain account. The new process improved the sigma value from <1 to ~4.

Mumbai, Maharashtra, IN

* Design and deployment of Ethernet and ATM based LAN solutions for customers.

Ms, Computer Science And Engineering

Be, Electronics And Communication