Transcript PI Meeting Presentation

Extreme Networking
Achieving Nonstop Network Operation
Under Extreme Operating Conditions
DARPA PI Meeting, July 23-26, 2002
Jon Turner
[email protected]
http://www.arl.wustl.edu/arl
Project Overview

Motivation
» data networks have become mission-critical resource
» networks often subject to extreme traffic conditions
» need to design networks for worst-case conditions
» technology advances making extreme defenses practical

Extreme network services
» Lightweight Flow Setup (LFS)
» Network Access Service (NAS)
» Reserved Tree Service (RTS)

Key router technology components
» Super-Scalable Packet Scheduling (SPS)
» Dynamic Queues with Auto-aggregation (DQA)
» Scalable Distributed Queueing (SDQ)
‹#› - Jonathan Turner - July 24, 2002
Prototype Extreme Router
Control
Processor
Field Programmable Port Ext.
Smart Port Card
Sys.
FPGA
ATM
Switch Core
FPX
FPX
FPX
FPX
Field
Programmable
SPC
SPC
SPC
SPC Port Extenders
Embedded
TI Processors
TI
Transmisson Interfaces
‹#› - Jonathan Turner - July 24, 2002
TI
TI
SDRAM
FPX
128
MB
Reprogrammable
SPC
Application
Device
TI
SRAM
4 MB
OPP
IPP
OPP
IPP
OPP
IPP
IPP
OPP
Cache
OPP
Pentium
APIC
IPP
IPP
North
Bridge
Switch Fabric
OPP
64
MB
FPX
Network
SPC
Interface
Device
TI
Resource Reservation in Internet?
Bandwidth reservation can provide dramatically
better performance for some applications.
 Obstacles to resource reservation in Internet.

» distaste for signaling protocols
» perceived complexity of IntServ+RSVP
» requires end-to-end deployment
» little motivation for service providers

How to get resource reservation in Internet?
» keep it simple
focus on top priorities - one-way unicast flows
 avoid complex signaling - leverage hardware routing mechanisms

» make it useful when only partially deployed
» provide motivation for ISPs to deploy it
‹#› - Jonathan Turner - July 24, 2002
Basic LFS Operation
Reserve
8 Mb/s to B
Reserve
bandwidth
A
Select best
next hop

Select path
and attempt
to reserve
10 Mb/s
available
5 Mb/s
available
20
Mb/s
5 Mb/s
available
available
2 Mb/s
available
Complete
reservation
20 Mb/s
available
B
Select path
and reserve
One way, unicast setup with partial reservation.
» complete reservations locally when bandwidth released
Optional ack returned by far-end access router.
 Reservation may terminate explicitly or time out.
 May alter reserved bandwidth but no re-routing.

‹#› - Jonathan Turner - July 24, 2002
Soft Reservations

Basic LFS provides firm reservations.
» user guaranteed bandwidth until releases

Can extend to provide soft reservations as well.
» soft reservation can be adjusted by the network as
traffic changes
» can be intermixed with firm reservations to provide a
firm minimum, plus more bandwidth as available

Uses of soft reservation.
» apps. that need guaranteed minimum and can sometimes
use more, but can adjust use to what’s available
» more rapidly responding congestion control for traditional
best-effort traffic
‹#› - Jonathan Turner - July 24, 2002
Basic IP Option for LFS


Code identifies LFS option.
Operations
» request firm reservation
» request soft reservation
» release state



code
Arate
Flags
»
»
»
»
»

IP header
(fixed part)
sender status request
sender network status request
public network status request
intra-domain status request
congested path
Rrate: requested rate.
Arate: allocated rate.
Trace used by each domain to
track usage.
‹#› - Jonathan Turner - July 24, 2002
length op. flags Rrate
trace
IP payload


Allocated rate stored at
“last hop” router for
status generation
F.P. rates with 4 bit
mantissa, 4 bit exponent.
» specify rates from 64 Kb/s
to 4 Gb/s , 6%
“granularity”
Use of Trace Field
acct. record
[A,B,..] thru X
A
X
acct. record
[A,B,..] thru Z
Y
domain U
X
Y
domain V
Z
domain W
Z
B
acct. record
[A,B,..] thru Y

Network providers need to monitor LFS usage for
network management and accounting purposes.
» trace field used by ingress router of each domain to mark
LFS packets with domain-specific identification
» egress router of each domain maintains record of each
LFS flow, including copy of trace field
» end-to-end records created through off-line accounting
resolution mechanisms
‹#› - Jonathan Turner - July 24, 2002
Status Reporting
sender status
sender net status
public net status
sender LAN
ISP U
ISP V
rcvr. LAN
intra-domain status
Basic LFS option supports sender status and trace
field for accounting.
 Network providers likely to want more.

» sender net status allows LFS service verification
» public net status allows “end-to-end” status check
» intra-domain status for verifying local status
» each “extra” status report requires insertion of
requestor’s IP address, increasing LFS option length
‹#› - Jonathan Turner - July 24, 2002
Partial Deployment

Receivers need not be LFS-aware.
» web site may use LFS to reserve bandwidth for streaming
media - users benefit, even without LFS-aware hosts

Issues with non-contiguous LFS domains.
» route changes may create “orphan reservations”
» no simple way to determine status reporter

No support for non-contiguous LFS domains.
» LFS router forwarding to a non-LFS router (or host)
strips LFS option and implements status reporting

status report includes IP address of reporting router, letting
sender know how far the reservation went
Public IP carrier can accept LFS option from client
networks (LAN) even if client net is not LFS-aware.
 Clients may use tunnel to access LFS service.

‹#› - Jonathan Turner - July 24, 2002
Regulating LFS Use - Net Access Svc
Permitting unconstrained access to LFS creates big
security vulnerability.
 Limit use to authorized users.
 Limit number of reservations and amount of
reserved bandwidth by authorized users.

» access router keeps record and enforces limits
» complication - user may use LFS from multiple locations


maintain records in distributed set of servers - each server keeps
records for some fraction of the users - use hashing to select
Access router needs means to identify user.
» host IP address insufficient (DHCP, NAT)
» encryption-based authentication (IPSEC)
Combine access control with usage accounting.
 What special issues arise with multiple domains?

‹#› - Jonathan Turner - July 24, 2002
LFS Video Demo Configuration
video
source
100 Mb/s
links
cross
traffic
sinks
video sink
cross
traffic
sources

Wavelet-coded video with and without LFS.
» competing datagram traffic
» with no reservation, lost packets cause poor video quality
» with reservation, high quality preserved
‹#› - Jonathan Turner - July 24, 2002
Video Demo - No Reservation
video flow - no reservation
datagram cross traffic flow 1
datagram cross traffic flow 2
video source
cross traffic sources
‹#› - Jonathan Turner - July 24, 2002
all sinks
Video Demo - With Reservation
video flow - with reservation
datagram cross traffic flow 1
datagram cross traffic flow 2
video sink
cross traffic sinks
‹#› - Jonathan Turner - July 24, 2002
Competing LFS Flows
flow 1 - no reservation
flow 2 - reservation added
flow 3 - no reservation
no reservations reservation for flow 2
sources
sink 2
sinks
‹#› - Jonathan Turner - July 24, 2002
sink 1
Partial Reservation
flow 1 - partial reservation made
source 1
flow 2
sink 1
‹#› - Jonathan Turner - July 24, 2002
sink 3
Completing Partial Reservation
flow 1 - completes partial reservation
flow 2 - drops reservation
sink 1
sink 3
‹#› - Jonathan Turner - July 24, 2002
Addition of Flow 3 Reservation
flow 3 - adds reservation
sink 3
‹#› - Jonathan Turner - July 24, 2002
sink 2
Performance of LFS at Single Link
Pareto distributed
session times make
little difference
1.E+00
1.E-02
m =50
Pr{delay >t }
1.E-04
1.E-08
very few flows 1.E-14
experience any
1.E-16
delay
0.00

OC-48 link can
carry 200 flows
of 12 Mb/s
200
1.E-10
1.E-12

100
1.E-06
400
load =.7
0.05
0.10
0.15
0.20
t (unit = average service time)
0.25
m = number of flows link can carry
exponential session times for flows, infinite queue
‹#› - Jonathan Turner - July 24, 2002
Sensitivity to Load and Hop Count
1.E+00
1.E-02
load =.9
.8
1.E-06
1.E-08
.7
1.E-10
1.E-12
.6
1.E-14
m =200
1.E-16
0.00
delay probability
scales linearly with
number of hops
0.05
0.10
0.15
0.20
t (unit = average service time)
at 90% load, less than 1
flow in 100 delayed more
than 12% of session time
0.25
Pr{delay >t }
Pr{delay >t }
1.E-04
1.E-05
load =.7
m =200
1.E-06
1.E-07
1.E-08
hops =1
1.E-09
4
16
1.E-10
1.E-11
1.E-12
0.00
0.05
0.10
0.15
0.20
t (unit = average service time)
‹#› - Jonathan Turner - July 24, 2002
0.25
Overload Performance
1.0
"carried load"
0.9
with no buffer most
sessions still
succeed
no buffer
0.8
m =50
100
200
400
0.7
0.6
infinite buffer
0.5
0.5
0.6
0.7
0.8
0.9
1.0
1.1
offered load
buffer reduces
rejection fraction
at low loads
1.E+00
infinite buffer
Fraction "Rejected"
with infinite buffer,
no sessions get small
delays (10%)
1.2
1.E-01
no buffer
1.E-02
1.E-03
m =50
100
200 400
1.E-04
0.50
0.60
0.70
0.80
0.90
offered load
‹#› - Jonathan Turner - July 24, 2002
1.00
1.10
1.20
Summary

LFS provides simple reservations for QoS.
» no complex signaling, wire speed setup
» limited deployment can be broadly beneficial
» support for usage monitoring & accounting gives network
providers a motivation to deploy service

Network access service for regulating usage.
» preliminary specification has been developed
» uses IPSEC for host/user authentication
Performance analysis, simulation study underway.
 Routing issues.

» evaluate QoS routing with multiple-choice forwarding
» link state distribution for inter-domain routing
» inter-domain routing policies
‹#› - Jonathan Turner - July 24, 2002