1 / 8

Server Switch

Server Switch. Carly Ho Sarah Alsulaiman. Programmability?. Commodity chips have limited programmability, not comparable to even general purpose CPUs FPGA:

tom
Download Presentation

Server Switch

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Server Switch Carly HoSarah Alsulaiman

  2. Programmability? • Commodity chips have limited programmability, not comparable to even general purpose CPUs • FPGA: • C to HDL (SystemC ), combine HLL with concurrency models to allow faster design cycles. (Cadence, Synopsys and Agility Design Solutions) • Flow to HDL, a graphical dataflow approach to high-level design entry. (Annapolis Micro Systems, Inc.'s CoreFire Design Suite and National Instruments LabVIEW)

  3. Best Programmable Device? • Microcontroller • Electronics Technology magazine awarded Stratix IV GX FPGA “Award for the best programmable device”

  4. Tradeoffs • Processing traffic: • Driver: higher performance, more effort to program and debug • User space: easy for development but sacrifice performance • Two step indirect lookup key extraction • Between extra ASIC logic and TCAM space

  5. Evaluation Issues • Most of the tests only compare one other system setup • The BCube network architecture they used was also built by many of the same paper writers—would it perform differently on different server architecture?

  6. Problems • In-network caching: • increases CPU usage but saves bandwidth • Cost of 14% CPU usage even when cache hit ratio = 0 • Extra delay to generate a QNC queue reply messages • User library based on Broadcom SDK, only on Linux and VxWorks • PCI-E latency • Field matching is limited by the HW

  7. Limitations • Limited hardware forwarding programmability • High packet processing latency due to switching chip to CPU communications • Restricted form factor and low speed • The internal bandwidth doesn’t match the total external one. Because they don’t think all traffic to be delivered to CPU

  8. Thank You!

More Related