LISA21 - Computing Performance: On the Horizon

Computing Performance: On the Horizon
Brendan Gregg
The chase for higher performance in computing is pervasive: it is the driving reason for many new technologies and a common feature of updates. While we can expect incremental performance improvements to our existing software and hardware (with Moore's law for processors a well-known example), it is harder to predict new technologies. This talk discusses the current performance improvements that you will likely be adopting for processors, memory, disks, networking, runtimes, hypervisors, and more, as well as discussing where things are headed with predictions for new technologies. The future of performance is increasingly cloud-based with hardware hypervisors and custom processors, meaningful observability of everything down to cycle stalls (even as cloud guests), and high-speed syscall-avoiding applications that use BPF, FPGAs, and io_uring.
View the full LISA21 program at www.usenix.org/conference/lis...

[@fanily4072]

Thanks a lot for putting this together. This is a wealth of good information at just the right level.

[@felixrabe]

Good stuff, good stuff, learning a lot here. Greetings from a fellow HNer.

[@YoloMonstaaa]

27:00 Regarding "adaptive runtime internals", ScyllaDB automatically benchmarks the systems drives to automatically tune runtime characteristics. It's a little off topic but maybe someone's interested in further reading about the topic.

[@DantalionNl]

Its unfortunate the comparison of new and promising processors missed the upcoming RISC-V architecture and the vendors that have started to create procesors with it (Western Digital, SiFive, Espressif)

[@tacticalcenter8658]

Seems like the smaller chips get, the more they can fit on a wafer. But often more chips on these smaller wafers will be defective. They would have to do the maths, but that may be acceptable but not idea. Yields might improve over time.

[@xiangzhang4935]

Impressive