{"id":170536,"date":"2010-08-22T16:39:46","date_gmt":"2010-08-22T16:39:46","guid":{"rendered":"https:\/\/www.microsoft.com\/en-us\/research\/project\/seawall\/"},"modified":"2019-08-19T15:24:47","modified_gmt":"2019-08-19T22:24:47","slug":"seawall","status":"publish","type":"msr-project","link":"https:\/\/www.microsoft.com\/en-us\/research\/project\/seawall\/","title":{"rendered":"Seawall"},"content":{"rendered":"

Seawall provides new ways to share the network in datacenters.<\/em><\/p>\n

SideCar<\/h1>\n

While investigating how we could get explicit feedback from the network middle to the ends for Seawall, we struck upon a way to provide programmability on a modest fraction of all packets flowing through the switches.\u00a0\u00a0Packets are re-directing, via sampling or marking, to\u00a0commodity servers that are directly attached to and\u00a0function as\u00a0dedicated programmable processors for these packets. We call this the SideCar <\/strong>architecture. Using SideCar, we show how many\u00a0novel datacenter problems can be solved. Refer to the HotNets paper for further details.<\/p>\n

<\/span><\/span><\/p>\n\t\t\t

\n\t\t\t\t\t\t\t
\n\t\t\t\t\t\n\t\t\t\t\t\t