Download presentation
Presentation is loading. Please wait.
1
SACRIO - An Active Buffer Mangement Scheme for Differentiaed Services Networks Saikrishnan Gopalakrishnan Cisco Systems Narasimha Reddy Texas A & M University
2
Outline Introduction Statement of Problem SACRIO - Approach and Implementation Results Summary
3
Introduction Diff-serv network - Markers and Droppers Scalable -- no per-flow state maintained in core routers RIO - RED with IN/OUT 3-drop precedence mechanism R/Y/G
4
Introduction(contd)
5
Introduction ( contd..) RIO - Core router recognizes IN/OUT Uses RED as buffer management Maintenance of per-flow state details are pushed to the edge routers IN/OUT packets are aggregated
6
Problems and Observation The achieved rates of flows with higher bandwidth < their target rates UDP will get unfair advantage when aggregated with TCP flows Lower the link’s apparent excess bandwidth closer the achieved rate with ideal target
7
Statement of Problem Achieved rates different from contract rates Difficulty in sharing excess bandwidth Can we do better without per-flow state?
8
Basic Ideas of SACRIO How to make the link function with no apparent excess bandwidth? –Apply Local Remarking of packets –IN and IN2 packets close to link capacity Convert OUT/IN into OUT/IN2/IN within the router Convert IN2 OUT on the output link How to do OUT IN2 conversion?
9
Local Remarking
11
SACRIO ( Contd..) Allow a target rate of OUT packets per flow –Convert packets into IN2 below this rate If Per-flow state is maintained, this can be done accurately Sampling and Caching to segregate the flows whose OUT i > target –Employ fixed amount of state transparently
12
Sampling and Caching Use fixed amount of transparent state Much like caches in memory systems –Transparent to the user/architecture –Improves performance –Allows scalable deployment of state –More state, more service –Provided service - a function of state and taffic
13
Sampling and Caching(contd) Sample flows to see if state needs to be maintained –State management policy Sampling allows scalability Cached flows can be regulated accurately –Resource management policy Provide aggregated service for other flows
14
Basis of SACRIO Local remarking of packets Employment of fixed amount of transparent state
15
SACRIO in Detail
16
Cost of SACRIO Per-packet handling cost = O(1) –Cached flows require an addition, division at the end of observation period, computation of OUT to IN2 conversion probability –Counters of total IN and OUT packets –Can be further reduced through sampling Memory cost = O(p), p = the state in the cache
17
SACRIO and 3-color markers 3-color markers are employed at the edge –Have no information of resource consumption within the network SACRIO employed inside the network, locally at each router
18
3-color marking & SACRIO
19
SACRIO Performance SACRIO does not explicitly drop packets that are not converted into IN2 – Local resource management policy controls this Different state management policies, resource management policies achieve different goals.
20
Performance goals Curtail per-flow OUT BW to preset limit Curtail per-flow OUT BW to fair share of excess BW ( C - IN)/ n –Requires estimation of n, the # of flows –Can we use the cache to estimate n –Cache hit ratios could provide an idea
21
Estimating # of flows If flows are equal rate, n = K/h –K = cache size, h = hit ratio Traffic is not uniform across flows Caches capture high-rate flows Estimate1 N’ = n(pinned) + m/h’ Estimate 2 N’’= (OUT/OUT(pinned) +1) N’
22
Simulations
23
UDP zero reservation
24
UDP non-zero reservation
25
Link with different loads
26
RTT bias reduction
27
Multiple Routers
28
Additive property
29
Complex Topology
31
When # of flows is not known
32
Effects of Under and over estimation of alpha
33
Effects of OP and RTT
34
SACRIO Scalability Seems to work with ns-2 simulations Do caches work in real networks? Analyzed Network traces at NLANR
35
SACRIO Scalability
37
Conclusion SACRIO employs Local remarking and limited transparent state SACRIO shown to curtail per-flow OUT BW consumption SACRIO is scalable based on analysis of internet traffic SACRIO is additive -better performance with more deployment
38
Future Work Build a Linux-based prototype router with limited state Demonstrate and verify the costs associated with fixed amount of state Use sampling and caching for other purposes such as traffic monitoring
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.