Food & Beverages

A peer-assisted server-based error recovery approach for IPTV networks

Description
A peer-assisted server-based error recovery approach for IPTV networks
Published
of 2
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Related Documents
Share
Transcript
  A Peer-Assisted Server-Based Error Recovery Approach for IPTVNetworks Aytac Azgin and Yucel Altunbasak School of Electrical and Computer EngineeringGeorgia Institute of Technology  Abstract —To achieve the desired service quality in IPTV networks,we need effective and resource-efficient error recovery techniques toovercome the problems encountered during the delivery of the IPTVcontent to the end-users. To enable quick recovery from packet losses,one idea is to utilize an error recovery server (ERServ) that is placedin-between the end-users and the content delivery server. Withinthis framework, error recovery process is initiated once an end-user sends a repair packet request to the ERServ, and the processis finalized when the repair packet transmitted by the ERServ inresponse is successfully delivered to the end-user. Even though the useof an ERServ presents an effective solution to improve the perceivedquality at the end-users, as the number of users connected to theERServ increases, we may observe scalability related problems (e.g.,dropped requests or increased response times). Consequently, duringsuch overloaded periods, we cannot guarantee the timely deliveryof the repair packets, which may lead to significant performancedegradations at the end-users. To prevent this from happening, wepropose to utilize peer-based error recovery in addition to server-based recovery, for which the main objective is to prevent the ERServfrom entering a non-responsive state. We achieve this objective byredirecting the error recovery load towards the end-users duringtimes when the ERServ reaches a critical server-load level. I. I NTRODUCTION Successful deployment of the IPTV service relies on thepromise of creating a reliable content delivery network that canprovide the subscribers with the desired quality of experience(QoE). However, the problems encountered during the delivery of the IPTV service make this a challenging task because of stringentservice quality requirements (e.g., packet loss and latency). Fur-thermore, since the subscribers are distributed over networks withvarying loss and delay characteristics, simultaneously accommo-dating the needs of each of these subscribers may cause inefficientutilization of the network resources, which may introduce stabilityproblems.To minimize these problems, various approaches have beenproposed to reduce the error recovery load in the network. Theseapproaches typically focused on utilizing dedicated error recoveryservers, which are placed in between the content delivery serverand the end-users, to help the users quickly recover from theirpacket losses. By using error recovery servers and retransmittingthe lost packets to end-users using a unicast-based approach, wecan also improve the effectiveness of multicast proactive repairstreams. However, because of the possibility of simultaneouslyserving thousands of subscribers, error recovery servers are notcapable of supporting these many users on a consistent basis. Tominimize the occurrence rate of server overload cases, peer-basedrecovery techniques are proposed to distribute the error recoveryload towards the user side (i.e., subscribers of the IPTV network).These approaches typically utilize the error recovery server on aneed-basis, i.e., error recovery server acted as a backup to peer-based recovery.The main disadvantage for approaches that rely on peer-basederror recovery is the additional latency incurred during the re-covery process. Since the repair packets are delivered from theend-users, the limitations observed along the last mile (e.g., highpropagation delay and low uplink bandwidth availability) willhave a direct impact on the latency performance. To overcomethis limitation while still taking advantage of the peer-basedrecovery process, our objective in this paper is to develop a server-based error recovery framework that only utilizes the peer-supportmechanism on a need-basis. Utilizing the error recovery server asthe initial source for the repair packets, we plan to increase theefficiency of the error recovery process. Furthermore, by utilizingsupport-peers, and thereby countering the main weakness of a pureserver-based recovery approach, i.e., limited number of users thatcan be supported by the same error recovery server, we also planto improve the reliability performance of the IPTV networks. Inour joint error recovery framework, we essentially utilize peer-based recovery during overload periods to minimize the numberof recovery requests, to which the server cannot respond in time.In the next section, we give an overview of the proposed approachand the method of analysis.II. P ROPOSED  A PPROACH The main objective of the proposed research is to develop ascalable server-based error recovery framework, for which thesession peers are used to assist with the IPTV recovery so that theservice quality constraints at the end-users are satisfied. Within theproposed framework, error recovery server (ERServ) acts as thesole coordinator to initiate the peer-based error recovery process.Because of the bandwidth limitation on the peers’ uplink channel,the peer selection process during the error recovery process needsto avoid overutilizing the session peers.The optimization problem is stated as follows: min Ω ers  (1) s.t.  Φ ν   < τ  ( ν  ) φ  , ∀ ν   ∈  N ω ν   < τ  ( ν  ) ω  , ∀ ν   ∈  N  where  N   represents the user set that is connected to the sameERServ,  Ω ers  represents the downstream load at ERServ; and foruser  ν  ,  Φ ν   represents the quality constraint ( e.g. , failed deliveryrate),  τ  ( ν  ) φ  represents the quality threshold ( e.g. , maximum allowedpacket loss rate),  ω ν   represents the upstream load reserved for thepeer-based recovery process, and  τ  ( ν  ) ω  represents the maximum The 8th Annual IEEE Consumer Communications and Networking Conference - Work in Progress (Short Papers) 978-1-4244-8790-5/11/$26.00 ©2011 IEEE527                                                     Fig. 1. Error recovery operation at ERServ. allowed upstream bandwidth utilization. In Figure 1, we illustratethe basic operation of the proposed error recovery process thatutilizes two operation modes:  peer-disabled recovery  mode, whichrepresents the server-based recovery operation, and  peer-enabled recovery  mode, which represents the joint server-based and peer-based recovery operation. The main design difference between thetwo approaches is the way the transmission queues are utilized tohandle the recovery requests. That is, ERServ is only responsiblefor sending repair packets for the peer-disabled scenario, whereasit also needs to send the peer-request packets for the peer-enabledscenario. By utilizing two queues for peer-enabled recovery, wecan separate the processing and transmission of request packetsfrom that of repair packets. Since a request packet is muchsmaller than a repair packet, a small portion of ERServ’s downlink bandwidth would be sufficient for timely delivery of peer-requestpackets. The ratio used for this bandwidth separation depends onthe server load after taking into account the impact of deliverydeadline for each of the received requests. This way, we can takeinto account  ( i )  the additional delay incurred using peer-basederror recovery and  ( ii )  the recovery deadline for requests comingfrom peer-disabled users.To decide when to make a transition from one recovery mode tothe other, we examine the state of the transmission queues. Eachrecovery mode is associated with certain servicing-thresholdsto check whether or not the transition conditions are achieved.Specifically, in peer-disabled recovery, whenever the repair queuelength exceeds the threshold  δ  H  , recovery mode changes to peer-enabled recovery. In peer-enabled recovery, all recovery requests(except for a specific subset of them) are forwarded as peer-requests. To determine the recovery requests that are handleddirectly by the ERServ, we use the information on the probabilityof meeting the recovery deadline. That is, if the probability of meeting the recovery deadline using peer-based recovery is belowa certain threshold and if the repair packet can be delivered earlierusing server-based recovery, then the ERServ forwards the requestto its repair packet transmission queue.To determine the load at the ERServ,  Ω ers , we use  ν  ∈ N   ρ ν  [ P   p − d ς  rp + P   p − e  P  τ,ers ς  rp +(1 − P  τ,ers ) ς  rq  ] , where  ρ ν  represents the packet loss rate for user  ν  ,  ς  rp  represents the server-based recovery overhead,  ς  rq  represents the peer-based recoveryoverhead,  P   p − d  (or  P   p − e 1 ) represents the probability of theERServ being in peer-disabled (or peer-enabled) mode, and  P  τ,ers represents the probability of using server-based recovery when theserver is in peer-enabled state. 1 Note that  P  p − d  = 1 − P  p − e . To approximate the equation for  P   p − d , we need to find theexpected duration of being in either of the recovery states. Assumethat the threshold for making a transition to the peer-enabledstate is given by  δ  H  . Then, the expected duration of being inpeer-disabled state,  E  [ T   p − d ] , becomes equal to the average timerequired for the queue size to hit  δ  H   the first time. In [1] theauthors examined the hitting time for the workload in an M/G/1queue. Since the transmission queue for the peer-disabled state canbe represented with an M/D/1 system, we can use the equationsdeveloped in [1] by assuming the workload threshold to be equalto  δ  H  /µ  p − d .To solve the equation for  E  [ T   p − d ] , we need to find the waitingtime distribution at the ERServ (see [2]). Then, to find theexpected duration of being in the peer-enabled state, we usea two-step procedure. The first step completes when the repairpacket queue length becomes  0  the first time. Then we startmonitoring the peer-request queue. The first time the peer-requestqueue length drops below  δ  L , we disable the peer-based recoverymode. To find the expected completion time for the first step, E  [ T  (1)  p − e ] , we use the settling time for the M/D/1 system (see[3]). The expected completion time for the second step,  E  [ T  (2)  p − e ] ,equals the time for the queue length to drop below the giventhreshold the first time, which can be found using the queuesize distribution and the servicing/arrival rates. Then, we can use E  [ T   p − d ] / ( E  [ T   p − d ] +  E  [ T  (1)  p − e ] +  E  [ T  (2)  p − e ])  to find  P   p − d .To find  P  τ,ers , we compare the delivery time of a repair packetunder both scenarios. If the additional delay of going through thepeers for error recovery reduces the probability of receiving therepair packet significantly, then the repair packets are transmittedby the ERServ. By varying the success threshold, we can vary thenumber of requests that can use server-based recovery in peer-enabled state. The delivery time for peer-based recovery dependson three factors: peer-assignment policy, group loss rate, andmulticast group size. We plan to use a weighted round robin basedpeer-assignment policy that does not prioritize the support peersbased on the forward-trip-times, unless we can ensure a successfuldelivery only by going through the closest-peer. Consequently, theequations developed for the transient M/D/1 queuing system canbe used to find the success rate and delay distribution for peer-based recovery and server-based recovery to make the optimaldecision.III. C ONCLUSION In this paper, we introduced a peer-assisted server-based errorrecovery framework to improve the scalability performance of IPTV networks, while achieving the desired reliability levels. Themethodology used to analyze the proposed recovery framework is also presented. The next step in our research is to integrate theequations developed for the recovery overhead at the ERServ intothe server modules designed for our simulation framework and todetermine the resource optimal thresholds.R EFERENCES[1] S. M. Ross and S. Seshadri, “Hitting time in an M/G/1 queue,”  Journal of  Applied Probability , vol. 36, pp. 934–940, 1999.[2] G. J. Franx, “A simple solution for the M/D/c waiting time distribution,” Operation Research Letters , vol. 29, pp. 221–229, 2001.[3] G. D. Stamoulis and J. N. Tsitsiklis, “On the settling time of the congestedG/G/1 queue,” 1989. 528
Search
Similar documents
View more...
Related Search
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks