Jonathan Turner - 10/25/99
Washington
WASHINGTON UNIVERSITY IN ST LOUIS
1
Design of a Flexible Open Platform for High Performance Active - - PowerPoint PPT Presentation
Design of a Flexible Open Platform for High Performance Active Networks Sumi Choi, Dan Decasper, John Dehart, Ralph Keller, John Lockwood, Jonathan Turner ,Tilman Wolf Washington University Applied Research Lab http://www.arl.wustl.edu/arl/
Jonathan Turner - 10/25/99
Washington
WASHINGTON UNIVERSITY IN ST LOUIS
1
Washington
WASHINGTON UNIVERSITY IN ST LOUIS
2 Jonathan Turner - 9/22/99
GTechnology advances adding new functionality to
» logic capabilities growing much faster than IO » packet classification, per flow queueing becoming common » single chip packet processing engines with 16 processors now becoming available
GApplication-specific processing in routers could
» active networking is one way to exploit trend » alternative model
–signalling and resource reservation –packet classification and flow-specific routing
GKey challenge is application software. GNeed better experimental platforms for researchers.
Washington
WASHINGTON UNIVERSITY IN ST LOUIS
3 Jonathan Turner - 9/22/99
GModular components.
» ability to swap components - both hardware and software
– routing, signalling, management software – address lookup and packet classification – queueing and packet scheduling
» open, documented and straightforward interfaces
GDynamic insertion of application-specific processing.
» active networking model and others
GHigh performance.
» gigabit links and scalability to large numbers of ports » packet processing rates of at least a million/ second per link » application-specific processing on large fraction of traffic » need credible demonstrations to influence commercial practice
Washington
WASHINGTON UNIVERSITY IN ST LOUIS
4 Jonathan Turner - 9/22/99
ANPEs IPPs LCs
Switch
LCs ANPEs OPPs CP AN Processing Element
APIC
Pentium Sys. FPGA 32-64 MB Cache NB
Input Port Processor
VCI VCI OUT
Control Processor
information needed by SPCs
processing
Washington
WASHINGTON UNIVERSITY IN ST LOUIS
5 Jonathan Turner - 9/22/99
ANPEs IPPs LCs
Switch
LCs ANPEs OPPs CP
3 4 6 4
Washington
WASHINGTON UNIVERSITY IN ST LOUIS
6 Jonathan Turner - 9/22/99
ANPEs IPPs LCs
Switch
LCs ANPEs OPPs CP
9 9 9
Sys. FPGA
APIC
Pentium Cache NB
2 6 2 6 2 6
Sys. FPGA
APIC
Pentium Cache NB
9 9 9
Washington
WASHINGTON UNIVERSITY IN ST LOUIS
7 Jonathan Turner - 9/22/99
G Std. proc. for “ plain” IP packets.
» classification & routing, header processing, output queueing
Packet Classification and Routing Driver SAPF Packet IPv4/6 Header Processing
Packet + Flow Id
Packet Scheduler Active Function Dispatcher Resource Controller Selector/ Dispatcher IP Packets Driver
Active Packets Plain Packets
Kernel Plugins . . .
. . . . . .
. . . . . . . . .
G Active packets move through configured kernel plugins.
» active function dispatcher passes packets to instances of plugin objects » instantiates objects or triggers download of plugin class, as needed » streamlined processing of SAPF packets using pre-established state
Washington
WASHINGTON UNIVERSITY IN ST LOUIS
8 Jonathan Turner - 9/22/99
anetd Java VM
ANTS
Active Plugin Loader
Plugin Requestor Plugin DB Controller Policy Controller Security Gateway
ANN Manager
Plugin DB Key DB
Routing & Signalling
Control Processor Kernel
Code Server Code Server
Active Code DB
Kernel
Active Plugin Loader Plugin Requestor Security & Policy Controller
Switch Fabric
Kernel
Active Plugin Loader Plugin Requestor Security & Policy Controller
Processing Elements
Policy Rules
Washington
WASHINGTON UNIVERSITY IN ST LOUIS
9 Jonathan Turner - 9/22/99
I O I O I O I O SE SE SE SE O I O I O I O I
Processor Module APIC System FPGA Connectors Optoelectronics Transmission Coder & Decoder Fiber jumper to front panel Main Circuit Board
Washington
WASHINGTON UNIVERSITY IN ST LOUIS
10 Jonathan Turner - 9/22/99
(250Kg+100Kb)
SDRAM
(32 MB)
SRAM
(1 MB)
Utopia Interface (3.2 Gb/s) Utopia Interface (3.2 Gb/s)
SRAM
(1 MB)
SDRAM
(32 MB)
64 100 MHz 32 32 64 100 MHz 6.4 Gb/s
Line Card PE
G Stackable port card
» can be combined with PE
G Programmable hardware
» FPGA technology » flexible memory config. » change on-the-fly
G Reprogrammable
Application Device (RAD)
» fully reprogrammable » four separate memory interfaces » memory bw: 2.4GB/ s
G Network Interface Device
(NID)
» relatively static » adapt for different line cards
G Variety of applications
» address lookup & packet class. » per flow queueing » traffic management » hardware plugins
Washington
WASHINGTON UNIVERSITY IN ST LOUIS
11 Jonathan Turner - 9/22/99
GHigh performance active networking need not be an
» scalable systems with gigabit links and terabit throughputs are possible with current/ near-term technology » on-going technology improvements will make AN economically viable
GNeed to focus on active application development. GNeed better abstractions, tools, APIs for developers. GEffective & open experimental platforms are essential.
» provide realistic testbed » provide more convincing demonstrations » enable system researchers and developers to build on each
Washington
WASHINGTON UNIVERSITY IN ST LOUIS
12 Jonathan Turner - 9/22/99