shivkumar kalyanaraman rensselaer polytechnic institute 1 edge-based traffic management building...
TRANSCRIPT
Shivkumar KalyanaramanRensselaer Polytechnic Institute
1
Edge-based Traffic Management Building Blocks
David Harrison, Shiv Kalyanaraman, Sthanu Ramakrishnan
Rensselaer Polytechnic Institute
http://www.ecse.rpi.edu/Homepages/shivkuma
I E
I
EI
E
Logical FIFO
B
Shivkumar KalyanaramanRensselaer Polytechnic Institute
2
Private Networks vs Public Networks QoS vs Congestion Control: the middle ground ?
Overlay Bandwidth Services: Key: deployment advantages A closed-loop QoS building block
Services: Better best-effort services, Assured services, Quasi-leased lines…
Overview
Shivkumar KalyanaramanRensselaer Polytechnic Institute
3
Motivation: Site-to-Site VPN Over a Multi-Provider Internetwork
International Linkor
International Linkor
Shivkumar KalyanaramanRensselaer Polytechnic Institute
4
Private networks over Public networks Can we reduce (not eliminate !) coordination requirements
for QoS deployment? Tolerate heterogeneity Incremental deployment Faster deployment cycles Dynamically provisioned services
Complexity Issues: Design: int-serv, RSVP, RTP … Implementation: diff-serv, CSFQ… Upgrades Configuration Management
Focus of this talk!
Shivkumar KalyanaramanRensselaer Polytechnic Institute
5
Problem: Inter-domain QoS Deployment Complexity
Solutions: Enable incrementally deployable edge-based QoS New closed-loop building blocks for efficiency Reduce (not eliminate!) coordination reqts. No upgrades!Tradeoffs: limited service spectrum
Today’s solutions require upgrade of multiple potential bottlenecks, and complex multi-provider coordination
Internetwork or WANWorkstationRouter
Router
RouterWorkstation
Router
Router
Shivkumar KalyanaramanRensselaer Polytechnic Institute
6
Our Model: Edge-based building blocks
New: Closed-loop control !Policy/Bandwidth Broker
I E
I
EI
E
Logical FIFO
B
Model: Inspired by diff-serv; Aim: further interior simplification
Shivkumar KalyanaramanRensselaer Polytechnic Institute
7
Closed-loop BB: Take-Home Ideas
FIFO
B
Loops: differentiate service on an RTT-by-RTT basis using edge-based policy configuration.
B
Priority/WFQ
Scheduler: differentiates service on a packet-by-packet basis
Shivkumar KalyanaramanRensselaer Polytechnic Institute
8
Queuing Behavior: Without Closed-loop Control
End system
Bottleneckqueue
Shivkumar KalyanaramanRensselaer Polytechnic Institute
9
Queuing Behavior: With Overlay Edge-Edge Control
edge devices
Results: efficient core operation, rate adaptation in O(RTT)
Shivkumar KalyanaramanRensselaer Polytechnic Institute
10
Edge-based Performance Customization Key idea: bottlenecks consolidated at edges, closer to application => incorporate application-awareness in QoS
Eg: L4-L7 aware buffer management. For TCP traffic: dramatically reduce timeouts:
Do not drop retransmissions or small window packets
Potential: application-level QoS services, active networking, edge-based diff-serv PHB emulation etc
Shivkumar KalyanaramanRensselaer Polytechnic Institute
11
Closed-loop Building Block Reqts
#1. Edge-to-edge overlay operation, #2. Robust stability #3. Bounded-buffer/zero-loss,
#4. Minimal configuration/upgrades + incremental deployment
#5. Rate-based operation: for bandwidth services
Not available in any congestion control scheme… Related work: NETBLT, TCP Vegas, Mo/Walrand, ATM
Rate/Credit approaches
Shivkumar KalyanaramanRensselaer Polytechnic Institute
12
Overlay Control: Concepts
Load: = i ; Capacity: ; Output Rates: i At all times: >= i (single bottleneck)
During congestion epochs, set i < i, Eg: i = min{i , i} Single bottleneck: Reverse queue growth within 1 RTT
Key: detect congestion: a) purely at the edges => overlay technology b) detect in a loss-less manner!
i
1
.
.
.
n
1
.
.
.
n
Shivkumar KalyanaramanRensselaer Polytechnic Institute
13
Implementation model
Overlay state: Edge-to-edge VL association at ingress and egress One token-bucket shaper (LBAP) per VL loop: Rate = i(t) ; burstiness:
Ingress Edge(Shapes edge-to-edge aggregate at rate i)
Egress Edge (feeds back measured rate i during congestion epochs)
Interior Node(modeled to identify congestion epochs)
i(t)
Ingress shaper
End-to-end traffic
ri(t)
Shivkumar KalyanaramanRensselaer Polytechnic Institute
14
Congestion Detection: Hypothetical Model
Mark all packets if the interior queue crosses N N is upper bound on transient burstiness during underload (I.e.
when i < )
If any marked packets seen by egress edge during measurement interval , i is fed back: begin epoch
If marked packets are not seen for a full interval , declare end of epoch and stop feedback of i.
Ingress Edge(Shapes edge-to-edge aggregateat rate i)
Egress Edge (feeds back measured rate i during congestion epochs)
Interior Node(helps identify congestion epochs)
Shivkumar KalyanaramanRensselaer Polytechnic Institute
15
Impln: Overlay Congestion Detection Emulate prior model, albeit without Interior assist
N: aggregate burstiness bound : per-VL accumulation bound.
Congestion epoch beginning: Measure per-VL accumulation qi = (i i) Per-VL accumulation qi exceeds 2 => epoch begins
Congestion epoch end: qi <= epoch ends Hysteresis helps ensure that queue drains
Shivkumar KalyanaramanRensselaer Polytechnic Institute
16
Increase/Decrease Dynamics Increase:
Additive increase of 1 pkt/ every interval ( >= RTT) Decrease:
i = min{i, i } every interval during the congestion epoch Properties (single bottleneck):
queue guaranteed to reduce within of feedback The lower rate is held till queue is drained
Input Rate Dynamics
i
time
i
can be set larger than 0.5
Shivkumar KalyanaramanRensselaer Polytechnic Institute
17
1
n n
1
+ -
+
1
n
)(ta n
Bottleneck { j }, Q ueue = q, Q ueue contribution of flow i = q
f
1
n
i F low { i}, Accum ulation = a , Input R ate = , O utput R ate =i i
D elay
j ji
i
Multi-bottleneck stability
Incremental drain provisioned for incremental accumulation Sum of input rates upper bounded; output rates lower bounded
Shivkumar KalyanaramanRensselaer Polytechnic Institute
18
Multiple Bottleneck FairnessThroughput versus Number of Bottlenecks
Linear Network: 1 flow (VL) crosses k bottlenecks. Each bottleneck has 4 cross flows (VLs).
Min-potentialdelay fairness
Proportionalfairness
Edge-to-edgeControl
Shivkumar KalyanaramanRensselaer Polytechnic Institute
19
Overlay Bandwidth Services
Basic Services: no admission control “Better” best-effort services Denial-of-service attack isolation support Weighted proportional/priority services
Advanced services: edge-based admission control Assured service emulation “Quasi-leased-line” service
Key: no upgrades; only configuration reqts…
Shivkumar KalyanaramanRensselaer Polytechnic Institute
21
Isolation of Denial of Service/Flooding
TCP starting at 0.0s UDP flood starting at 5.0s
Shivkumar KalyanaramanRensselaer Polytechnic Institute
22
Backoff Differentiation Policy:
Backoff little (as) when below assurance (a), Backoff (as) same as best effort when above assurance (a) Backoff differentiation quicker than increase differentiation
Service could be potentially oversubscribed (like frame-relay) Unsatisfied assurances just use heavier weight.
Edge-based Assured Service Emulation
1 > AS >BE >> 0
r =r + min(r, AS aa
if no congestion
if congestion
Shivkumar KalyanaramanRensselaer Polytechnic Institute
23
Bandwidth Assurances
Flow 1 with 4 Mbps assured + 3 Mbps best effort
Flow 2 with 3 Mbps best effort
Shivkumar KalyanaramanRensselaer Polytechnic Institute
24
Assume admission control and route-pinning (MPLS LSPs). Provide bandwidth guarantee. Key: No delay or jitter guarantees!
Adaptation in O(RTT) timescales Average delay can be managed by limiting total and per-
VL allocations (managed delay) Policy:
Quasi-Leased Line (QLL)
1 > BE >> 0
r =r + if no congestion
if congestionmax(aaa
Shivkumar KalyanaramanRensselaer Polytechnic Institute
25
Quasi-Leased Line Example
Background QLL starts with rate 50Mbps
Best-effort VL quickly adapts to new rate.
Best-effort rate limit versus time
Best-effort VL starts at t=0 and fully utilizes 100 Mbps bottleneck.
Shivkumar KalyanaramanRensselaer Polytechnic Institute
26
Quasi-Leased Line Example (cont)
Bottleneck queue versus time
Starting QLL incurs backlog.
Unlike TCP, VL traffic trunks backoff without requiring loss and without bottleneck assistance.
Requires more buffers: larger max queue
Shivkumar KalyanaramanRensselaer Polytechnic Institute
27
Quasi-Leased Line (cont.)
Worst-case queue vs Fraction of capacity for QLLs
Single bottleneck analysis:
q < b
1-bB/w-delay products
For b=.5, q=1 bw-rtt
Simulated QLL w/edge-to-edge control.
Shivkumar KalyanaramanRensselaer Polytechnic Institute
28
Signaling/Configuration Issues Simple: Each edge-box independently sets up loops only with
other edges it intends to communicate Address-prefix list based configuration for VPN application Minimal overhead to maintain the loop: a leaky bucket, 8-
bytes every 250 ms or so of overhead
ISP configures ONE separate class at potential bottlenecks for overlay controlled traffic
Scalable to inter-domain VPNs as long as each edge does not have to manage > 100s of loops
Properties: Bounded scalability, simplified interior configuration, incremental deployment, simple set of overlay services.
Shivkumar KalyanaramanRensselaer Polytechnic Institute
29
Edge-to-Edge Principle ? Tradeoff between public and private network philosophies: Private network characteristics:
Differentiated Svcs, simple forms of overlay QoS Bounded scalability and heterogeneity
Edge-to-edge loops, queue bounds, policy/BB scalability, bridging approach to inter-domain QoS
Public network characteristics: Incremental deployment. O(1) complexity. Stateless interior inter-network Minimal interior upgrades, configuration support. Use of robust, stable closed-loop control for efficiency and
adaptation in O(RTT) timescales.
Shivkumar KalyanaramanRensselaer Polytechnic Institute
30
Current Work With bottlenecks consolidated at the edge:
What diff-serv PHBs or remote scheduler functionalities can be emulated from the edge ?
What is the impact of congestion control properties and rate of convergence on attainable set of services ?
Areas: Application-level QoS: edge-to-end problem Dynamic (short-term) services Congestion-sensitive pricing: congestion info at the edge
Edge-based contracting/bidding frameworks Point-to-set svcs: more economic value than pt-to-pt svcs
Dynamic provisioning for statistical muxing gains
Shivkumar KalyanaramanRensselaer Polytechnic Institute
31
Summary
Private Networks vs Public Networks QoS vs Congestion Control vs Throwing bandwidth
QoS DeploymentL Simplified overlay QoS architecture
Intangibles: deployment, configuration advantages Edge-based Building Blocks & Overlay services:
A closed-loop QoS building block Basic services, advanced services