Proposal of an IP QoS framework:

Transcription

1 Proposal of an IP QoS framework: Scalable Services - ScalServ Jaime Dias, Manuel Ricardo ScalServ - Classes of Service Application real-time elastic Best Effort () reference service. Majority of the applications Assured Delivery (AD) Real time applications; applications demanding minimum bandwidth Don t guarantee a maximum end-to-end delay No packet loss in router queues loss intolerant AD loss tolerant interactive Packet transit delay equals transmission + propagation times (controlled queue lengths) Resources explicitly reserved using end-to-end signalling protocol (E2ESP) Interactive bulk assynchronous Less than Best Effort (L) best effort service with lower priority. L packets transferred only after or AD packets; residual bandwidth reserved to avoid L starvation. L 2º Seminário RTCM / Coimbra 2 1

2 Code Points Service class has a code point (SSCP); marked in packets Code Points are global Remarking due to congestion inexistence of reservation (AD) impossibility of guarantee AD service (AD NG-AD). ScalServ field: 2 bits Description SSCP Best Effort 00 Less than Best Effort 01 Assured Delivery 10 Non-Guaranteed Assured Delivery a) DS Field SS Field b) DS Field SS Field ECN Field 2º Seminário RTCM / Coimbra 3 Architecture Nodes classified based on QoS granularity: Fine Granularity () Coarse Granularity () nodes B adequate for networks having complex resource management LAN, ad-hoc networks nodes Adequate processing an high number of flows Core networks HB B domain B B domain B B B domain B domain 2º Seminário RTCM / Coimbra 4 2

3 Architecture nodes process AD packets per flow AD signalling and L packets per class B nodes process all packets per class do not process AD signalling domain B HB B B B B domain domain B B domain 2º Seminário RTCM / Coimbra 5 End-to-end Signalling protocol (E2ESP) Sender host router router Receiver host Receiver-initiated Send packets PATH RESV PATH RESV Sender host router router Receiver host Sender-initiated PATH+RESV PATH+RESV Send packets 2º Seminário RTCM / Coimbra 6 3

4 Data Plane Classification Class filter Flow filter Shaping r1+r2, b1+b2 rn, bn Scheduling L AD p 2º Seminário RTCM / Coimbra 7 Data Plane Shaping Classification Class filter AD class only Flow filter Discrete shaping Shaping r1+r2, b1+b2 rn, bn control over individual flows control of the maximum shaping delay per flow implemented by the source host and the first next router (policing) Scheduling L AD Aggregate shaping p less control over each flow lower average delay per aggregate 2º Seminário RTCM / Coimbra 8 4

5 Data Plane Scheduling Classification Class filter Flow filter Shaping r1+r2, b1+b2 rn, bn Scheduling L AD While (1){ if (ADQueue not empty) transmit ADpacket; else if (LQueue not empty and (Queue empty or Lcredit>= Lpacket.length)){ transmit Lpacket ; Lcredit -= Lpacket.length; if (Lcredit<0) Lcredit= 0; } else if (Queue not empt y) { transmit packet ; if (LQueue not empt y) Lcredit+ = packet.length*lresidualrate; } } p 2º Seminário RTCM / Coimbra 9 Data Plane Interior Node Classification Classification Classification Class filter Class filter Scheduling L AD Scheduling L AD Scheduling L AD p p p a) b) c) More resources (over-provisioning), less differentiation 2º Seminário RTCM / Coimbra 10 5

6 IP tunnelling Two modes: Leased Line adequate for VPNs Class mapping adequate for IPv4-IPv6 transition tunnels e MIP tunnels destination source protocol ID ScalServ code point Payload destination source protocol ID ScalServ code point Payload Inner packet destination source protocol ID ScalServ code point = AD Payload destination source protocol ID ScalServ code point = inner SSCP Payload Outer packet a) Leased line mode b) Class mapping mode source host tunnel end-point tunnel end-point destination host PATH+RESV (1) PATH+RESV (4) PATH+RESV (5) (inner IP layer) (8) (7) (6) PATH+RESV (2) tunnel (outer IP layer) (3) 2º Seminário RTCM / Coimbra 11 Mobile IP HA HB CN B B domain domain HB AR1 HB AR2 B AR3 B MN 2º Seminário RTCM / Coimbra 12 6

7 P+R P+R UPDATE P+R Mobile IP Tolerant AD forwarding forwards AD packets with no reservation. AD packets without reservation pass through a default shaper (r 0, b 0 ) AD packets of rejected reservation are discarded No guarantees that AD packets continue receiving the same service Moderate AD service during handoff, while not starving or L traffic. E2ESP end-node address update Some nodes in the new path are the same. nodes common to old and new paths only update the Filter SPEC with the new CoA address. P+R CN RELEASE CN P+R UPDATE CN P+R P+R RELEASE P+R UPDATE P+R MN MN MN handoff (1) (2) (2) 2º Seminário RTCM / Coimbra 13 Simulations /L differentiation enables interactive and background applications to share the same link, using work-conserving scheduling AD/ differentiation allows real-time applications to share the same link with bursty applications, providing QoS to real-time (AD) flows Discrete shaping provides a firm control of the maximum reshaping delay per AD flow; aggregate shaping provides low average reshaping delays per aggregate 2º Seminário RTCM / Coimbra 14 7

8 Simulations /L differentiation HTTP server HTTP client Scheduler 512 kbit/s FTP server FTP client 2º Seminário RTCM / Coimbra 15 Simulations /L differentiation http, bytes received (kbyte) http alone with ftp with ftp L (3%) time (s) Scenario Flow Response time (ms) Throughput (kbit/s) Expected Verified Expected Verified HTTP alone HTTP (100%) 512 (100%) FTP HTTP HTTP (+70,0%) (61,7%) with FTP FTP (38,3%) HTTP HTTP 3866 (+3%) 3843 (+2,5%) 497 (97%) 500 (97,7%) with FTP L FTP (3%) 12 (2,3%) 2º Seminário RTCM / Coimbra 16 8

9 Simulations AD/ differentiation (12 kbit/s, 90 bytes) audio (video-tel.) (64 kbit/s, 1500 bytes) video (video-tel.) (64 kbit/s, 40 bytes) audio (str.) (256 kbit/s, 1100 bytes) video (str.) Scheduler 512 kbit/s recv audio (video-tel.) recv video (video-tel.) recv audio (str.) recv video (str.) http1 server http1 client http2 server http2 client 2º Seminário RTCM / Coimbra 17 Simulations AD/ differentiation audio () video () audio (AD) video (AD) max. AD delay max. delay delay (ms) ,5 1 1,5 2 2,5 3 3,5 time (s) Scheduling delay of video-telephony flows without and with AD/ differentiation 2º Seminário RTCM / Coimbra 18 9

10 96 Simulations AD/ differentiation audio () video () 96 audio (AD) video (AD) rate (kbit/s) rate (kbit/s) cumulative loss (kbyte) time (s) time (s) Throughput of video-telephony flows without and with AD/ differentiation audio () video () time (s) Packet loss of video-telephony flows without AD/ differentiation cumulative bytes lost/bytes sent (%) 55% 50% 45% 40% 35% 30% 25% 20% 15% 10% 5% 0% time (s) audio () video () 2º Seminário RTCM / Coimbra 19 Simulations AD/ differentiation Scenario Flat AD/ Flow Scheduling delay (ms) Throughput (kbit/s) Packet loss Max. Avg. Max. Avg. Min. Avg. Audio (tel.) 468,3 289,1 14,0 11,8 6,0 1,8% Video (tel.) 468,4 267,3 69,0 51,1 24,8 20,0% Audio (str.) 468,7 290,0 127,6 63,5 4,7 0,7% Video (str.) 468,2 290,9 308,6 241,6 100,0 5,5% Audio (tel.) AD 48,2 11,1 12,4 12,0 10,5 0,0% Video (tel.) AD 46,9 31,4 64,4 64,0 55,9 0,0% Audio (str.) 569,9 333,0 209,2 63,8 4,6 0,2% Video (str.) 550,3 324,8 324,3 235,5 98,0 7,8% 2º Seminário RTCM / Coimbra 20 10

11 Simulations Discrete and aggregate shaping (12 kbit/s, 90 bytes) audio (jitter = 10 ms) Link1 recv audio shaping Scheduler 512 kbit/s video Link2 (jitter = 50 ms) recv video (64 kbit/s, 1500 bytes) 2º Seminário RTCM / Coimbra 21 Simulations Discrete and aggregate shaping 25 Audio shaping 50 Video shaping 20 before shaping after discrete shaping 40 after aggregate shaping delay (ms) delay (ms) before shaping after discrete shaping after aggregate shaping packet number packet number Scenario Flow Link jitter (ms) Max. delay (ms) Average delay (ms) Discrete shaping Audio Video Aggregate shaping Audio Video º Seminário RTCM / Coimbra 22 11

12 Conclusions Scalable Services a replacement for current best-effort service. Scalable term used with two meanings: 1. networks can be added/enlarged without compromising the services offered 2. the framework applies to local, access and core networks; from an end-to-end perspective 3 global service classes: AD, and L expected to serve all type of applications. Simulation results proved that 1. discrete and aggregate shaping is valid, and 2. the proposed scheduler is simple, effective and contributes for a good network efficiency. 2º Seminário RTCM / Coimbra 23 Future Work E2ESP specification Definition of a resource management model for domains Definition of business models Definition of a mobile IP framework with ScalServ and IP micromobility Deployment of ScalServ in ad-hoc networks with low resources; ex.: Body Area Networks (BAN) and Personal Area Networks (PAN). 2º Seminário RTCM / Coimbra 24 12

13 Commercial/Industrial Interests IP + ScalServ = Always-On-with-QoS Networks: o Heterogeneous L2 networks o Ad-hoc networks (PAN) o Corporative networks o ISP, Operator networks Terminals: o Desktop, laptop, o Mobile Phone, PDA Effective and simple end-to-end QoS deployment Easy to understand + Simple to deploy Wide-spreading of multimedia services higher revenues +L Solution for ISPs overloaded core networks due to P2P applications 2º Seminário RTCM / Coimbra 25 Thanks! Questions? 2º Seminário RTCM / Coimbra 26 13

14 Transition to ScalServ A ScalServ region is a set of one or more contiguous ScalServ domains. IP packets passing through a non-scalserv region shall be marked, when entering a ScalServ region, as non-guaranteed (SSCP=11) When a node receives an E2ESP signalling packet from a non- ScalServ region it must indicates the reservation as non-guaranteed The acceptance of non-guaranteed AD flows is let to applications 2º Seminário RTCM / Coimbra 27 Transition to ScalServ 1st fase B networks B 3rd fase 2nd fase B B HB B B B B domain B domain B 2º Seminário RTCM / Coimbra 28 14