BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Disk|Crypt|Net: rethinking the stack for high-performance video st
 reaming - Ilias Marinos (Computer Lab)
DTSTART:20170727T140000Z
DTEND:20170727T150000Z
UID:TALK74991@talks.cam.ac.uk
CONTACT:Liang Wang
DESCRIPTION:Conventional operating systems used for video streaming employ
  an in-memory disk buffer cache to mask the high latency and low throughpu
 t of disks. However\, data from Netflix servers show that this cache has a
  low hit rate\, so does little to improve throughput. Latency is not the p
 roblem it once was either\, due to PCIe-attached flash storage. With memor
 y bandwidth increasingly becoming a bottleneck for video servers\, especia
 lly when end-to-end encryption is considered\, we revisit the interaction 
 between storage and networking for video streaming servers in pursuit of h
 igher performance. We show how to build high-performance userspace network
  services that saturate existing hardware while serving data directly from
  disks\, with no need for a traditional disk buffer cache. Employing netma
 p\, and developing a new diskmap service\, which provides safe high-perfor
 mance userspace direct I/O access to NVMe devices\, we amortize system ove
 rheads by utilizing efficient batching of outstanding I/O requests\, proce
 ss-to-completion\, and zerocopy operation. We demonstrate how a buffer-cac
 he-free design is not only practical\, but required in order to achieve ef
 ficient use of memory bandwidth on contemporary microarchitectures. Minimi
 zing latency between DMA and CPU access by integrating storage and TCP con
 trol loops allows many operations to access only the last-level cache rath
 er than bottlenecking on memory bandwidth. We illustrate the power of this
  design by building Atlas\, a video streaming web server that outperforms 
 state-of-the-art configurations\, and achieves ~72Gbps of plaintext or enc
 rypted network traffic using a fraction of the available CPU cores on comm
 odity hardware.\n\n\nBio:\n\nIlias is a PhD student at the Computer Labora
 tory\, under the supervision of Robert Watson and Mark Handley. His resear
 ch interests include operating systems and high-performance network/storag
 e stacks.
LOCATION:FW26\, Computer Laboratory\, William Gates Building
END:VEVENT
END:VCALENDAR
