<html><body><div style="font-family: garamond,new york,times,serif; font-size: 12pt; color: #000000"><div></div><div data-marker="__QUOTED_TEXT__"><p style="margin-left: 50px;" data-mce-style="margin-left: 50px;">
</p><p style="margin-left: 50px;" data-mce-style="margin-left: 50px;">Muhammad Shahbaz will present his FPO, " <!--StartFragment-->Enabling Programmable Infrastructure for Multi-tenant Data Centers<!--EndFragment--> " on Friday, 9/7/2018 at 11am in CS 302.</p><p style="margin-left: 50px;" data-mce-style="margin-left: 50px;">The members of his committee are as follows: Adviser: Nick Feamster; Readers: Jennifer Rexford and Ben Pfaff (VMware Inc.); Examiners: Wyatt Lloyd, Michael Freedman, and Nick Feamster.</p><p></p><p style="margin-left: 50px;" data-mce-style="margin-left: 50px;"></p><p style="margin-left: 50px;" data-mce-style="margin-left: 50px;">Everyone is welcome to attend. A copy of his thesis is available in CS 310. Abstract follows below.</p><p style="margin-left: 50px;" data-mce-style="margin-left: 50px;"><br data-mce-bogus="1"></p><p style="margin-left: 50px;" data-mce-style="margin-left: 50px;">Today’s data centers are large, shared infrastructures hosting hundreds of thousands</p><p style="margin-left: 50px;" data-mce-style="margin-left: 50px;"></p><p></p><p style="margin-left: 50px;" data-mce-style="margin-left: 50px;">of tenants over a vast network of servers. Operating these data centers frequently</p><p style="margin-left: 50px;" data-mce-style="margin-left: 50px;">entails incorporating new infrastructure services that require customizing the</p><p style="margin-left: 50px;" data-mce-style="margin-left: 50px;">behavior of its switches and exploiting unique characteristics of data centers to scale.</p><p style="margin-left: 50px;" data-mce-style="margin-left: 50px;">Emerging programmable switch ASICs allow network operators to customize the</p><p style="margin-left: 50px;" data-mce-style="margin-left: 50px;">behavior of physical switches. Yet, virtual switches running on servers in multitenant</p><p style="margin-left: 50px;" data-mce-style="margin-left: 50px;">data centers are still fixed function and composed of large, complex software</p><p style="margin-left: 50px;" data-mce-style="margin-left: 50px;">codebases. Modifying these switches requires both intimate knowledge of the switch</p><p style="margin-left: 50px;" data-mce-style="margin-left: 50px;">codebase and extensive expertise in network protocol design, raising the bar for customizing</p><p style="margin-left: 50px;" data-mce-style="margin-left: 50px;">these switches prohibitively high.</p><p style="margin-left: 50px;" data-mce-style="margin-left: 50px;">In this dissertation, we address these challenges by, first, presenting the design and</p><p style="margin-left: 50px;" data-mce-style="margin-left: 50px;">implementation of PISCES: a programmable, protocol-independent software switch</p><p style="margin-left: 50px;" data-mce-style="margin-left: 50px;">derived from Open vSwitch (OVS), a fixed-function hypervisor switch, whose behavior</p><p style="margin-left: 50px;" data-mce-style="margin-left: 50px;">is customized using P4. PISCES is not tethered to specific protocols; this</p><p style="margin-left: 50px;" data-mce-style="margin-left: 50px;">independence makes it easy to add new features. We also show how the compiler can</p><p style="margin-left: 50px;" data-mce-style="margin-left: 50px;">analyze the high-level specification to optimize forwarding performance. Our evaluation</p><p style="margin-left: 50px;" data-mce-style="margin-left: 50px;">shows that PISCES performs comparably to OVS and that P4 programs for</p><p style="margin-left: 50px;" data-mce-style="margin-left: 50px;">PISCES are about 40 times shorter than equivalent changes to OVS source code.</p><p style="margin-left: 50px;" data-mce-style="margin-left: 50px;">Next, we demonstrate how such programmable switches help build scalable infrastructure</p><p style="margin-left: 50px;" data-mce-style="margin-left: 50px;">services by exploiting the unique characteristics of data-center networks. We</p><p style="margin-left: 50px;" data-mce-style="margin-left: 50px;">use these switches to address the multicast scalability problem in multi-tenant data</p><p style="margin-left: 50px;" data-mce-style="margin-left: 50px;">centers and present the design of an infrastructure service, Elmo, that scales multicast</p><p style="margin-left: 50px;" data-mce-style="margin-left: 50px;">by taking advantage of the data-center characteristics; specifically, the symmetric</p><p style="margin-left: 50px;" data-mce-style="margin-left: 50px;">topology and short paths in a data center. In Elmo, a PISCES switch encodes multicast</p><p style="margin-left: 50px;" data-mce-style="margin-left: 50px;">group information inside packets themselves, reducing the need to store the</p><p style="margin-left: 50px;" data-mce-style="margin-left: 50px;">same information in hardware switches, which instead read the encoded information</p><p style="margin-left: 50px;" data-mce-style="margin-left: 50px;">to route packets to recipients. In a three-tier data-center topology with 27,000 hosts,</p><p style="margin-left: 50px;" data-mce-style="margin-left: 50px;">Elmo supports a million multicast groups using a 325-byte packet header, requiring</p><p style="margin-left: 50px;" data-mce-style="margin-left: 50px;">as few as 1,100 multicast group-table entries on average in hardware switches, with a traffic overhead as low as 5% over ideal multicast.</p></div></div></body></html>