Title: Diffserv deployment in the wide area: network design and testing
1Diffserv deployment in the wide area network
design and testing
- Tiziana Ferrari
- Italian National Institute for Nuclear Physics
- ferrari_at_cnaf.infn.it
- TF-TANT task force
- http//www.dante.net/tf-tant/
2Overview
- Why diffserv?
- Test programme
- testbed
- interim results
- classification and marking
- policing
- scheduling
- QoS measurement
- services
- VLL
3Motivation
- Why diffserv?
- independent of layer 2 technologies
- interoperability
- diffserv domains ? National Research Networks
(NRN) - diffserv region NRN 1 diffserv core domain
- independent diffserv implementation in each
diffserv domain - traffic aggregation
- scalability
- transparency to the application
4Motivation (cont)
- (some) benefits
- network layer
- VLL service, a potential replacement of ATM
permanent connections - resource allocation on congested links
- QoS support in/to NRN which are not ATM based
- application layer
- QoS support for mission critical applications
(RSVP - diffserv) - e.g. good QoS support to client-server
distributed applications thanks to traffic
aggregation
5Diffserv and the European network model
DS domain
DS domain
MPLS
NRN
NRN
Non DS capable domain
NRN
Marking policing
scheduling
TEN-155
DS domain
NRN
(marking)
shaping
NRN
DS domain
DS domain
6Overview of test programme
- Diffserv
- DSCP class selectors, EF, AF
- test of basic QoS features in diffserv platforms
- parameter tuning
- interoperability
- services definition, implementation, validation
through test applications (video streaming, IP
telephony, distributed applications) - diffserv and intserv (end-to-end QoS)
- diffserv and MPLS
- QoS measurement
- Policy management
7Work accomplished (June 99-Oct 99)
- network configuration
- baseline performance testing
- marking
- policing
- functionality, token bucket parameter tuning
(normal and exceed burst size) with TCP traffic - CB-WFQ (Class Based WFQ) traffic isolation
- scheduling
- CB-WFQ traffic isolation, VLL service in the
wide area - SCFQ (Self Clocked Fair Queuing) EF, AF and
best-effort
8Work accomplished (cont)
- QoS measurement (ongoing)
- parameters one-way delay and one-way delay
variation - LAN and WAN (VLL service implementation and
validation) - traffic generators with GPS antennas
- NTP
- definition of test services
- point to point virtual leased line
- one-to-many virtual leased line
- resource allocation on congested links
- better than best-effort
- services for remote instrumentation control (AF)
9Diffserv test network
Uni. Bologna
CSELT
EPFL
10LAN layout (example)
11Equipment
- Test workstations (ATM, Ethernet, Fast Ethernet,
Giga Ethernet interfaces) - traffic generators (3 SmartBits, Netcom Systems)
connected to GPS rx - ATM switch(es) in each test sites
- 2 Cabletron Ethernet switches
- DS capable routers
- CISCO C7200 and/or C7500
- IBM 2212 and IBM 2216 (in 5 sites, 2 routers per
site)
12Interim Results
13Classification, marking and policing
- Multi-field classification OK
- packet marking OK
- testing of different exceed actions OK
- policing (at a configurable rate)
- token bucket parameter tuning for TCP traffic
performance optimisation - normal burst
- exceed burst
14Policing test of exceed actions
drop
precedence 0
Throughput SWITCH 1.20 Mbps DANTE 0.38
Mbps (drop effect)
15Policing TCP burst tolerance
- Normal and exceed burst tuning with TCP traffic
- single and multiple TCP streams
- optimum values are functions of the rate R at
which - traffic is policed, in particular
- normal burst 0.5 sec R
- independently of the platform
- exceed burst 2 normal burst
16Policing TCP burst tolerance (cont)
- CISCO platforms
- Target rate 1.25 Mbps
17Policing TCP burst tolerance (cont)
- IBM platforms (IBM 2212)
- Small TCP burst tolerance in a policer can
completely starve a TCP stream. - Token bucket depth is function of the rate at
which traffic is policed - ? tuning needed
18Policing TCP burst tolerance (cont)
EF behaviour aggregate of TCP traffic, target
rate 163 Kbps
(IBM 2212)
19Policing, two-parameter TBoptimum burst size
for different traffic rates
IBM 2212
- Small TCP burst tolerance it can completely
starve a TCP stream - optimum token bucket size function of the
policing rate
20Scheduling and policingtwo-parameter TB and
premium buffer size
Recommendation token bucket size lt premium
buffer size (for low packet loss probability)
21Scheduling CB-WFQ
- Scheduling implementation to provide minimum
bandwidth - guarantees to classes
- tested on CISCO platforms - C7200, IOS 12.0(5)T -
- class definition based on class selectors for
aggregation ? OK - bandwidth allocation no starvation, dynamic
allocation in case of classes which are not fully
deploying the capacity guaranteed to them ? OK - traffic isolation between EF BA and best-effort
- several combinations of TCP and UDP traffic
- variable number of TCP streams (different
aggregation degrees)
22CB-WFQ set-up
23CB-WFQ traffic isolation, 1 TCP stream
- Inconsistent results, for some sites TCP
throughput lt 1.25 Mbps - different ATM burst tolerances on the WAN
connections seem seem to be - the cause ? traffic isolation on WAN
connections can be an issue
24CB-WFQ traffic isolation, multiple TCP streams
- Overall TCP performance of the behaviour
aggregate improves when the - aggregation degree increases
- TCP performance of a single connection still low
- ? problem under investigation
25Scheduling Self Clocked Fair Queuing
- IBM 2212
- Policy (traffic profile, validity period,
diffserv action) - Diffserv action (type of marking, queue type,
bandwidth) - optimum traffic isolation (tests only with UDP)
Premium 163.8 Kbps guaranteed (8 PPP link
bw) Assured 819.2 Kbps (40 PPP bw)
26QoS measurement
- Resource allocation monitoring
- for resource allocation and network dimensioning
- performance measurement
- passive
- active (invasive traffic)
- for service validation
- to understand the effect on end-to-end
performance of buffering in one router or in a
chain
27Parameters
- parameters of interest
- one-way delay, requirements clock
synchronisation - GPS based synchronisation
- SmartBits (Netcom Systems)
- NTP (Network Time Protocol)
- NTP client / server hierarchy
- one-way delay variation
- deployment of inter packet gap at rx and tx
- packet loss probability
- throughput
28QoS measurement 1-way delay and instantaneous
delay variation (LAN)
Goal VLL service implementation and
validation Metric Type-P-One-way-Delay (RFC
2679), type-P-one-way-ipdv Variables
- WFQ/SCFQ EF queue depth
- packet size
- EF queue weight
- traffic volume
29One-way delay average vs packet size
30Latency in time
31ipdv (without best-effort)
32ipdv (with best-effort)
33Services
- Virtual leased line
- point to point
- one to many
- capacity allocation on congested links (e.g. US
links) - better-than-best-effort (qualitative definition)
34VLL
35VLL implementation
36Future work
- Weighted Random Early Discard (WRED) testing
- completion of IBM testing
- study of TCP related issues ACK colouring, TCP
and UDP aggregations with AF - interoperability testing (other future platforms
NORTEL, TELEBIT, TORRENT, Linux) - service implementation and validation
- diffserv and MPLS
- diffserv and intserv
- policy management and bandwidth brokerage
37To know more
- Activities of the TF-TANT task force
- http//www.dante.net/tf-tant
- Diffserv test programme and interim test results
- http//www.cnaf.infn.it/ferrari/tfng/ds
- QoS measurement test programme
- http//www.cnaf.infn.it/ferrari/tfng/qosmon