Transcript PowerPoint Presentation - Non Blocking Crossbar Switch

An introduction to the group and
its projects
Tony McGregor
[email protected]
WAND
Projects
•
•
•
•
•
•
•
•
•
CRCNet
Active Measurement
IP Measurement protocol
Passive Measurement
Simulation
Integrated measurement and simulation
Emulation Network
Physical layer switch
IPv6
• topology, mobile stacks, fast handover
• NZNOG ‘04
CRCNet
Introduction
• Project started almost 2 years
ago
• Rural communities were
frustrated by low speed
unreliable Internet access
• Develop a new platform suitable
to deploy future generation
(>>10Mbps) wireless networks
in rural and remote areas
• based around a mesh architecture
• Funded by Foundation for
Research Science and
Technology
CRCNet
Architecture
CRCNet
Stage 1 – Build Trial Network
Range of equipment
• 2.4Ghz (802.11b and g)
• Orinoco radio cards and APs
• Advantech and Soekris Biscuit PC
• Linksys wireless Ethernet bridges
• 5.8 GHz
• Proxim Quick bridge20
• Trango
Current Topology
CRCNet
Pirongia Site
CRCNet
HSK Site
CRCNet
MFR Site
CRCNet
Web Casting
• Between Hamilton Zoo and the
Fieldays site
• 6 wireless links
CRCNet
Stage Two – Platform Design
• Routing protocols for mesh networks
• Link Layer Design
• Design of a new node
AMP
Introduction
• NLANR’s active measurement project
• Approx 140 monitors, mostly in the USA.
• International deployments
•
•
a single AMP monitor in about a dozen other countries
some national AMPs (Australia, Taiwan, Russia soon)
• Measure
•
•
•
•
RTT
loss
topology
throughput (on demand)
• NSF funded
AMP
USA Sites
AMP
Architecture
Test Results
Web
browser
Active
Monitor
Analysis
machine
test traffic
(amp)
Test Results
Cichlid
Analysis
machine
(volt)
Active
Monitor
Other
target
Active
Monitor
AMP
Demo
AMP
Demo
AMP
Demo
AMP
Demo
AMP
Demo
AMP
Demo
AMP
Demo
AMP
Demo
AMP
Cost vs Function
• Design
•
•
•
•
dedicated machines
1ms accuracy
No GPS/CDMA
1 sample per minute
• Benefits
•
•
•
easy and cheap => wide deployment
full mesh
manageable
• Limits
•
•
no one-way delays (bidirectional traceroute, IPMP OWD)
very short events missed
AMP
Management
my SQL
d atab se
AMP
Vo lt
sy stem man ager
ALL
AMP
HPW REN
amp -k iwi
amp -p alo mar
mo n ito rs
AMP
New Zealand
• Beginnings of a New Zealand AMP mesh
•
•
•
•
Waikato
Auckland
APE
Ihug (offer)
Can fund more monitors and maintenance
•
•
need hosts (here?)
hosts provide space, power and network
IPMP
Introduction
• Current active measurement protocols have
weaknesses
•
•
multiple packets (overhead, phantom routes)
measurement of components (reverse path, CPU)
• IPMP combines path and delay measurement in a
single packet exchange with low router overhead
IPMP
Architecture
High perfomance ISP
IPM P
IPMP Enabled
router
IPM P
Measurement
Host
Non-IPMP Router
IPM P
Peering point
IPM P
Target Host
Progression of packet through the network
header
header
header
header
no change
Packet that leaves
measurement host
(one path record)
Path record added
at first IPMP
enabled router
Packet as it leaves
the kernel on the
target host
IPMP
Protocol (IPv4)
0
0
0
8
1
6
Version
Queue Type
00000000
Type
2
4
3
1
Checksum
Returned TTL
Length
Return Type
Path Pointer
(optional) data
(optional) Path Records
Padding (if required)
IPMP
Timestamps
• Router can use any timestamp it has available
• Resolving to real-time is not done in the packet
forwarding critical path
• Uses a separate packet exchange (information
request/reply)
•
•
supplies real-time reference points
other router information
IPMP
Information Reply
0
0
0
8
1
6
Version
00000000
00000000
Type
2
4
3
1
Checksum
00000000
Length
Precision
Performance Data Pointer
Forwarding IP Address
Accuracy
IPMP Processing Overhead
(optional) Path Records
(optional) performance data
IPMP
Uses
• POM made better
•
•
•
•
•
•
•
combined path and latency, no phantom routes etc
lower overhead
kernel based timestamps
explicit clock information
forward and reverse traceroute
DoS resistant
associates router interfaces
• One way delay from NTP
• Bandwidth Estimation
• Deployment (AMP, CRCnet)
Passive Measurement
Overview
• To support simulation work the group developed
passive header capture hardware.
• Known as Dag cards
• Speeds from Ethernet to OC48 (2.5Gbps WAN)
• Spun off a startup
•
•
•
Endace (www.endace.com)
now OC192
better support
Passive Measurement
Dag Overview
• Capture IP headers or full packet
• Add accurate timestamp
•
GPS or CDMA for external time
• Originally header trace focused
•
•
real-time flow based
security applications
• Optical splitter, electrical card relay or electrical tap
Passive Measurement
Dag 3 block diagram
Passive
Dag 4.2
Passive
WITS Traffic Archive
• Long traces from Auckland University and NZIX
•traces up to 45 days (3.2 billion packets)
•IP headers
•GPS timestamps
• Some analysis online
• Can fetch traces from NLANR
• Summary CD
Simulation
Introduction
• ATM-TN based
•
•
•
•
University of Calgary/Waikato partnership
parallel
BSDLite network stack (sort of)
high bandwidth delay, mixed real-time/TCP
• NS-2 with FreeBSD stack
•
•
new work
network cradle
• 802.11b link layer
Simulation
Example –TCP splitting
international channel
NZ Proxy
NZ Internet
US Proxy
US Internet
Web Clients
US Servers
Simulation
The simulation process
Simulation
Parameters
HTTP
Log
Logfile
generator
Digested
Logfile
Hostfile
generator
Host
List
buffer and
MSS info
Packet
trace
Simulator
HTTP Page
Latencies
Line and
Buffer use
Host
Information
Summarise
and
Plot
Graphs
query on
host
live
hosts
Internet
Pre-process
Simulate
Post-process
Simulation
Example –TCP spliting, Network parameters
• Bandwidth 34.369Mbps (E3)
• Delay
60ms
• TCP buffer size
•
•
proxy
servers
• MSS
• US delay
• NZ delay
32767 bytes
as measured
as measured
as measured
not simulated
Simulation
TCP Splitting – a single connection
Simulation
Introduction
Simulation
Introduction
Messim
Introduction
• Simulation is only accessible to very large network
operators and users
• AIM: Make simulation available to medium sized
enterprises
• Integrate measurement and simulation
• FRST funded
Messim
Introduction
Monitored Network
topology discovery
measurement
and analysis
alerts and
query results
workload
Workload
Model
Network
Model
validation
simulator
Messim
Projects
• Topology discovery
•
automated discovery of link layer devices
• Traffic Models
•
•
•
•
•
•
further development of specific models (e.g. peer to peer)
generic
Extraction of simulation parameters from traces
Extended range of network stack models
Continuous validation
Hardware flows analysis
Messim
Kernel space
User space
Network Stack Cradle
Mozilla / Bash / KDE / etc.
Network stack
FreeBSD 5 kernel
Messim
Network Stack Cradle
Network
User space
Simulator
Cradle (~200 functions)
Network stack
Messim
Generic models
• 2d Empirical distribution
Messim
Generic models
Messim
Generic models
Messim
Generic models
• Use WEKA machine learning algorithms to
•
•
cluster
classify
• For each cluster
•
•
simplify the rule set into terms for a network manager
produce an empirical distribution for each
• Allow simulations with different proportions of traffic
Emulation Network
Introduction
• There is a need for a structured environment in
which to build networks in the laboratory
•
•
validation of simulations
testing on network equipment
• The emulation network is two racks of PCs that can
be configured as
•
•
•
routers
end hosts
delay
• Plus configuration and measurement support
Emulation Network
Overview
Configure
PC
PC
PC
PC
PC
PC
PC
PC
PC
PC
R
R
SW
Ixia
Patch
Panel
Dag
H
Configure
R
R
H
delay
R
R
R
H
Monitor
(DAG)
Emulation Network
Usage and development
• Usage
•
•
•
•
Is a public facility
Has been used to debug AT switch
Used network trace capture and replay then Ixia script
Ihug traffic shaper
Bandwidth estimator
• Development
•
Physical layer switch
Crossbar Switch
Introduction
•
•
•
•
64 Port FastEthernet Crossbar switch
Fast / Flexible Reconfiguration
Link Monitoring
Latency Control
• Bandwidth limiting
• Self Documenting Network Topology
• Centralised Control
Crossbar Switch
Block Diagram – Overview
DaughterBoard
Uplink
DaughterBoard
Mainboard
DaughterBoard
12.8Gb/s
DaughterBoard
3.2Gb/s
Mainboard
• Crossbar
• Latency
• Bandwidth Limiting
Daughterboards
• Ethernet Interface
• Time Division MUX
Crossbar Switch
Block Diagram –Mother board
Uplink
DaughterBoard
12.8Gb/s
FPGA
FLASH
CPU
SDRAM
DaughterBoard
DaughterBoard
DaughterBoard
DDR SDRAM
(8GB max).
Crossbar Switch
PHY
Ethernet Ports
Block Diagram – Daughter board
Uplink to Motherboard
3.2Gb/s
PHY
Ethernet Ports
FPGA
Crossbar Switch
Daughter board Layout
Daughterboard Layout
Skamper
Overview
• Skitter for IPv6
• Hope to capture the growth of the IPv6 internet
IPv6 Stacks
Overview
• Small devices
•
•
•
•
•
•
One of the motivators for IPv6 is to provide addresses and other
support for small devices
a.k.a. cell phones
implementing a stack for embedded devices
little ram
moderate CPU speeds
prototype hardware development
• Fast handover between cells
•
•
normally may exceed 2s
reduce to around 150ms, l2 triggers, L3 preparation for handover
and timing improvements in protocols
NZNOG
Conference
• The New Zealand Network Operators Group has an
•
•
•
•
•
annual conference
The next one will be hosted by WAND
Jan 29-30 2004, at Waikato
Discounted registration (free?) for students
Hope to have a number of partial travel grants for
students
Could hold a parallel Academic Networking
Conference
•
need feedback