High performance computing using AVX-512 & Distributed Tracing

Date: 
Wednesday, March 21, 2018 - 18:00
Source: 
GoSF
Attendees: 
292
City: 
San Francisco

Agenda

6:00 Networking | Food | Drink

6:30 Speakers

• Talk 1: High Performance Computing in Go – Using 512-bit Advanced Vector Extensions (AVX-512)
• Talk 2: Navigating Distributed Traces

8:00 End

–––––––––––––––––––––––––––

Uber's mission is to bring transportation — for everyone, everywhere.

–––––––––––––––––––––––––––

Talk 1: Using 512-bit Advanced Vector Extensions (AVX-512) in Go (Frank Wessels, Minio)
Frank Wessels from Minio will talk about making use of AVX-512 in Go in order to dramatically increase performance for special compute needs. AVX-512 are 512-bit extensions to the 256-bit Advanced Vector Extensions SIMD instructions for x86 instruction set architecture (ISA) and is supported in Intel's Xeon Phi x200 (Knights Landing) and Skylake-X Core i7 and i9 models.

In particular, the talk will address the following subjects and improvement areas:

• How to add support for AVX512 on Skylake CPUs in Golang Assembly
• How to accelerate SHA256 performance by up to 8x for use in eg. fingerprinting and blockchain
• How to accelerate Reed Solomon encoding by 4x

About the Speaker
Frank Wessels is an entrepreneur in the IT industry. He started in 1993 in ISG Technologies as a developer for medical systems in Toronto, Canada. He then co-founded and served as chief architect for a venture developing a PACS system for radiologists. He also founded 3mensio Medical Imaging which focused on 3D visualization using GPU technology. In 2014 he switched to cloud computing and co-founding a company that built technology for high-speed transfer content over the internet. Lately, he has focused on Golang as well as become involved in object storage via a startup called Minio.

–––––––––––––––––––––––––––

Talk 2: Navigating Distributed Traces
Description: Building off Jaeger's open source distributed tracing system, we designed XYS to allow for interactive exploration of inter-service latency histograms and to directly connect long tail latency metrics to example traces.

About the Speaker:
Bill Westlin is a performance engineer at Uber where he has been working with distributed tracing to provide insight and introspection into the microservice architecture since 2016. He previously designed and built distributed databases for EMC and graduated from Carnegie Mellon University in 2013