Enhancing Interest Forwarding for Fast Recovery from Unanticipated Data Access Failure in NDN

Xiaoyan Hu＊,Xuhui Liu,Lixia Zhao,Jian GongGuang Cheng,2,3

1 School of Cyber Science ＆ Engineering,Southeast University,Nanjing 211189,China

2 Key Laboratory of Computer Network and Information Integration of Ministry of Education,Southeast University,Nanjing,China

3 Key Laboratory of Computer Network Technology of Jiangsu Province,Nanjing,China

Abstract: We show that an aggregated Interest in Named Data Networking (NDN) may fail to retrieve desired data since the Interest previously sent upstream for the same content is judged as a duplicate one and then dropped by an upstream node due to its multipath forwarding.Furthermore,we propose NDRUDAF,a NACK based mechanism that enhances the Interest forwarding and enables Detection and fast Recovery from such Unanticipated Data Access Failure.In the NDN enhanced with NDRUDAF,the router that aggregates the Interest detects such unanticipated data access failure based on a negative acknowledgement from the upstream node that judges the Interest as a duplicate one.Then the router retransmits the Interest as soon as possible on behalf of the requester whose Interest is aggregated to fast recover from the data access failure.We qualitatively and quantitatively analyze the performance of the NDN enhanced with our proposed NDRUDAF and compare it with that of the present NDN.Our experimental results validate that NDRUDAF improves the system performance in case of such unanticipated data access failure in terms of data access delay and network resource utilization efficiency at routers.

Keywords: named data networking; interest aggregation; multipath forwarding; data access failure; negative acknowledgement

I.INTRODUCTION

User generated data sharing and multimedia content delivery are more and more popular.Named Data Networking (NDN) [1],a promising future Internet architecture,has been proposed to fundamentally meet users’ growing demand on such content delivery and improve the content delivery performance of the Internet.NDN mainly focuses on (1) enabling access to content by name rather than original location,(2) offering content-based security rather than connection-based security,and (3) exploitingInterest aggregation,multipath forwarding,and in-network cachingto improve the network performance.

NDN addresses and routes every packet by name.NDN’s communication is driven by a requester sending an Interest packet which specifies the name of desired content.NDN network nodes1NDN node and router are interchangeably used in the rest of the paper.forward the Interest towards potential source(s) of its matching Data packet according to its name.An intermediate node aggregates multiple Interests for the same Data packet from different requesters and forwards only the first arrival upstream to solicit the matching Data packet (i.e.,Interest aggregation or Interest collapse).Interest aggregation reduces the transmission of redundant Interest and Data packets and makes use of network resources more efficiently.Any network node that has the matching Data packet carrying the desired content can reply to the Interest since each NDN Data packet is protected by a signature of its producer and can be verified by any network node (i.e.,content-based security).The matching Data packet returns via the path that the Interest travels along and is optionally cached by routers on the path (i.e.,in-network caching).In addition,NDN intrinsically supportsmultipath forwardingof an Interest,which aims to bring Data packets back to its requester(s) via multiple paths and then make NDN’s data transmission more robust and efficient.

However,in Section III,we will present that when multipath forwarding and Interest aggregation are simultaneously adopted in the forwarding of an Interest,in certain case,the aggregated Interest may fail to bring desired Data packet back to its requester even if the desired Data packet is available and the Interest is correctly forwarded by the network.Such unanticipated data access failure leads to delayed data access.To the best of our knowledge,we are the first that identify such data access failure.We further propose NDRUDAF,the mechanism that enhances the Interest forwarding in NDN to identify and fast recover from such unanticipated data access failure based on a negative acknowledgement (i.e.,NACK) [2,3].

The remainder of this paper is organized as follows.Section II gives the overview of NDN’s forwarding plane and analyzes the related work of correcting Interest forwarding in NDN.Section III describes the unanticipated data access failure due to the simultaneous adoption of Interest aggregation and multipath forwarding.NDRUDAF,the NACK based detection and fast recovery from the unanticipated data access failure is presented in section IV.The performance of the NDN enhanced with NDRUDAF is qualitatively analyzed in Section V.Section VI conducts an experimental study on the performance of NDRUDAF.Finally,we concludes our work in Section VII.

II.BACKGROUND AND RELATED WORK

2.1 Overview of NDN’s forwarding plane

This subsection briefly introduces NDN with a focus on its stateful forwarding plane.NDN is a receiver-driven and content-centric communication protocol.Its communication is driven by a requester sending an Interest packet which specifies the name of desired content and a nonce value that distinguishes the Interest from others with the same name.NDN replies to the Interest with a Data packet under the matching name and any node holding the matching Data packet can satisfy the Interest.

NDN nodes process an Interest according to the name in the packet.Upon the arrival of an Interest,an NDN node first looks up its Content Store (CS) to see if the requested Data packet is locally cached.If it is cached,the desired Data packet returns back to the requester via the face from which the Interest arrives.Otherwise,the node further consults its Pending Interest Table (PIT) to see whether the request for the desired content has already been sent upstream.If a matching PIT entry is found,the node compares the nonce value of the forwarded Interest recorded in the PIT entry with that of the new arrival Interest.If they are the same,the new arrival Interest is judged as a duplicate or loop one and dropped resulting in no data return on the path from which it arrives.Such duplication or loop detection reduces redundant traffic delivery in NDN.If they are different,the new arrival Interest is aggregated (Interest aggregation),would not be forwarded and waits for the return of the desired content.The nonce value of the new arrival Interest and the face from which the Interest arrives are recorded in the matching PIT entry.If there is no matching PIT entry,the node consults its For-warding Information Base (FIB) to determine how to forward the Interest.If a matching FIB entry is found,the Interest is forwarded accordingly.NDN naturally supportsmultipath forwarding.The matching FIB entry may offer several next hops and the Interest can be forwarded to the faces listed in the FIB entry.A PIT entry is created for the forwarding of the new arrival Interest recording its name,nonce,arrival face and outgoing face(s).If no matching FIB entry is found,the node discards the Interest as it does not know how to forward the Interest.When the Interest arrives at any node that holds the desired Data packet,the Data packet returns to the requester via the reverse path traversed by the Interest based on the arrival faces recorded in the matching PIT entries of these on-path nodes,and is optionally cached by these on-path nodes.The matching PIT entries in these on-path nodes are either removed when the matching Data packet returns or time out if no matching Data packet returns.

2.2 Related work

Interest aggregation can reduce redundant traffic transmission and improve the efficiency of the network.Dabirmoghaddamet al.[4] presented a mathematical model for characterizing a CCN router with a CS and a PIT and introduced an iterative algorithm for analyzing a hierarchical network of content routers in terms of the cache hit ratio at the CS,the Interest aggregation probability at the PIT,as well as the router response time at the object level of granularity.Numerical evaluations of the proposed model under realistic assumptions together implied that a small fraction of Interests in the system actually benefit from Interest aggregation.

However,Interest aggregation may also bring some negative influence on network performance in some scenarios.Garcia-Luna-Aceveset al.[5,6] illustrated that the NDN forwarding strategy is not safe when Interest loops occur,and there is no safe forwarding strategy that supports Interest aggregation and can detect Interest loops only by uniquely identifying Interests.They showed that an Interest loop occurs when one or more Interests requesting the same content are forwarded and aggregated by routers along a cycle.More specifically,each router forwards an Interest to its neighbor with the highest ranking among all the neighbors that can bring back the desired content.As each router ranks its neighbors independently,there may be long-term routing loops and routers in a routing loop form a cycle.Then a router in a cycle forwards an Interest and the Interest is forwarded along the cycle.Before receiving the Interest,another router in the cycle sends an Interest for the same content and the Interest is also forwarded along the cycle.Finally,the two Interests would be aggregated at the two routers,leading to an Interest loop and no data back.They proposed SIFAH which can prevent and detect such Interest loops when Interests aggregation is adopted.In SIFAH,routers and Interests are required to store the corresponding hop counts to the desired content respectively.A router forwards or aggregates an Interest from a neighbor only if the hop count stated in the Interest is greater than that from the router to the content through at least one another neighbor according to its FIB.SIFAH avoids Interest loops,but the Interest may not be forwarded via the path with the best performance.

Garcia-Luna-Aceveset al.[7] also presented a similar approach,CCN-ELF,to prevent undetected Interest loops without requiring any modification of NDN packet formats.In CCN-ELF,each FIB stores the distances from its neighbors to the desired data.A router also maintains a link-cost table,which stores the cost of the link from the current router to each of its neighbors.A router forwards or aggregates an Interest from a neighbor only if the distance from the neighbor to the desired content is greater than that from itself to the desired content via another neighbor.

This paper focuses on a different type of unanticipated data access failure.In our proposed scenario,the Interest aggregated by a matching PIT entry fails to retrieve the desired content even if the requested content is reachable and the Interest is successfully transmitted in the network since the aggregating Interest is forwarded via multiple paths and judged as a duplicate one by an upstream node.We introduce NDRUDAF to fast detect and recover from such data access failure.

III.THE FORMATION OF THE UNANTICIPATED DATA ACCESS FAILURE

3.1 Problem formalization

We first state how an unanticipated data access failure occurs due to the simultaneous adoption of Interest aggregation and multipath forwarding.When an Interest is forwarded via multiple paths,it is possible that some forwarding paths intersect at a certain router.Then the router will receive the same Interest from different paths one after another.According to the rule of Interest process in NDN,the router would forward the the first arrival Interest upstream and recognize the others arriving later as duplicates.Then when the matching Data packet returns,the router forwards the Data only through the path that the first arrival Interest travels along.As a consequence,the downstream routers on the other paths will not get a response.The PIT entries for the Interest at these downstream routers keep pending until they expire.If the Interest for the same Data issued by another requester arrives at one of these downstream routers before the PIT entries expire,Interest aggregation occurs aiming to reduce redundant Interest forwarding.However,as the desired Data packet will not pass the router that aggregates the latter Interest,the Interest aggregation not only fails to show it’s superiority,but also fails to retrieve the desired Data packet for the latter requester.In this case,unanticipated data access failure takes place.If the latter requester still desires the same Data,it will have to retransmit the Interest after the local PIT entry for the Interest expires.

3.2 An illustrative example

Figure 1 illustrates an example of such data access failure.The arrowheads in the figure indicate the next hops to content advertised by nodePaccording to the FIB entries stored in routers.Dashed lines indicate the transmission paths of Interests.The time when an event occurs at a router is indicated bytiandt1＜ t2＜ t3＜ t4＜ t5.ConsumerC1 sends an Interest for Data provided by nodePand the Interest arrives at the first hop routerR0 at the time oft1.The Interest is forwarded towards nodePby routerR0 via two pathsR1-R2-R3-R7-PandR4-R5-R6-R7-Pand arrives at routersR1 andR4 at the time oft2.After routerR4 forwards the Interest to routerR5,another Interest from consumerC2 for the same Data provided by nodeParrives at routerR4 at the time oft3 and is aggregated byR4.Then the Interest fromC1 via the pathR1-R2-R3-R7 arrives at routerR7 at the time oft4 and is forwarded to the content providerP.Later the Interest fromC1 via the other pathR4-R5-R6-R7 arrives at routerR7 at the time oft5 and is recognized as a duplicate one and discarded by routerR7.Then the requested Data packet returns toC1 via the pathP-R7-R3-R2-R1-R0 and no Data returns on the pathPR7-R6-R5-R4-R0.And therefore the Interest fromC2 fails to bring Data back toC2.If it still wants the Data,C2 needs to re-express the Interest after the matching PIT entry atC2 times out.That said,the data access atC2 is delayed anyway.

Fig.1.Unanticipated data access failure at consumer C2.

IV.NACK BASED DETECTION ＆ FAST RECOVERY FROM THE UNANTICIPATED DATA ACCESS FAILURE

This section describes our proposal,NDRUDAF,a NACK based mechanism that enhances the Interest forwarding and enables Detection and fast Recovery from such Unanticipated Data Access Failure.As indicated by the example in Figure 1,even though the desired Data packet is available and reachable and the Interest is correctly forwarded by routers,the simultaneous adoption of Interest aggregation and multipath forwarding sometimes leads to unanticipated data access failure.Such data access failure delays the data access at the consumerC2.The consumerC2 would retransmit the Interest after its PIT entry for the Interest times out if it still wants the Data packet.All the en-route nodes to the data source need to process the retransmitted Interest and create a PIT entry for the Interest again,which wastes network resources at these nodes.We argue that if the en-route node which aggregates the Interest can detect such data access failure and then forwards the aggregated Interest upstream again to solicit the desired Data packet,the delayed time would be less and network resources would be utilized more efficiently.

Algorithm 1.The processing of a NACK Interest with code “Duplicate” at a network node.Input:NACK(cname,nonce,iface) PitEntry←PIT.find(cname) if PitEntry≡φor PitEntry.RetryTimer expired or nonce∉PitEntry.N onceList then Stop processing else Forward the NACK downstream to the incoming face face of the Interest with the nonce value nonce PitEntry.NonceList-={nonce} PitEntry.IncomingFaces-={face} if PitEntry.NonceList = 0 then Remove PitEntry else if All outgoing faces of PitEntry return NACK without the desired data back then Pick nonce' from PitEntry.N onceList Prepare the Interest Int with cname and nonce' Forward Int upstream to solicit the desired Data end if end if end if

Yiet al.[2] and Zhanget al.[3] introduced a new Interest NACK to enable NDN routers to perform quick and informed recovery from network problems.NDRUDAF is based on the Interest NACK.When recognizing an Interest as a duplicate or loop one,an upstream network node forwards an Interest NACK that carries the same name and nonce as the original Interest,plus a NACK code “Duplicate” downstream.Upon the arrival of a NACK Interest with code “Duplicate”,if the NACK is a valid one,i.e.,a matching PIT entry is found and not expired,a network node would check the matching PIT entry to see whether another Interest is aggregated by the Interest.If there exists an Interest aggregation,the node would not demote the face from which the NACK arrives.Instead,it forwards the NACK Interest downstream only to the incoming face of the Interest recognized as a duplicate one by upstream nodes and removes the nonce and the arrival face of the duplicate Interest from the matching PIT entry.Then if the desired Data packet would not return from other outgoing faces recorded in the matching PIT entry,it tries to retrieve the desired Data packet again by forwarding the aggregated Interest to the face from which the NACK arrives and that has a higher probability of bringing back the desired Data packet.If the desired Data packet is indeed available and reachable,the requester whose Interest is aggregated would receive the Data packet soon without knowing that it has ever been involved in a data access failure and a NACK Interest is ever sent for its data retrieval.Our proposal NDRUDAF is sketched in Algorithm 1.Note that if there are multiple aggregated Interests and the aggregated Interest resent upstream is recognized as a duplicate one again,which should occur with a fairy small probability,the above process for NACK can be repeatedly taken at the aggregating router as long as the matching PIT entry does not expire.The desired Data packet returns to the requesters if any Interest sent upstream successfully brings back the desired Data packet.

V.DISCUSSIONS ＆ ANALYSIS

5.1 Performance implications

NDRUDAF improves the system performance in case of the unanticipated data access failure in terms of data access delay and network resource utilization efficiency at routers.We compare the system performance of NDN enhanced with NDRUDAF with that of NDN.The node that aggregates the Interest detects the data access failure via the returned NACK Interest with code “Duplicate”.Then it forwards the aggregated Interest upstream to solicit the desired Data packet without requiring the requester whose Interest is aggregated to wait for its matching PIT entry to time out or the receipt of a NACK Interest.Therefore,the retransmitted Interest has more chance to get cache hit before the desired cached Data packet that has just been delivered is replaced and the average data access delay would be less.A small average data access delay is essential for many popular applications today,e.g.,AR [8,9],real-time streaming [10].Besides,the en-route nodes on the whose Interest is aggregated,e.g.,the nodesR8 andR9 in path from the node that aggregates Interests to the requester Figure 1,need not to remove their matching PIT entries after the PIT entries time out nor create corresponding PIT entries again when the retransmitted Interest arrives,which saves the process resources at these nodes.Moreover,the average PIT size and the average pending time of PIT entries at routers can be reduced as compared to that in NDN.

5.2 Storage complexity

NDRUDAF is pretty straightforward,invokes no additional space cost,and only requires NDN routers to check whether a NACK Interest with code “Duplicate” aggregates other Interest(s) as compared to that in NDN.As for the storage complexity,a non-negligible space can be saved if NDN is enhanced with NDRUDAF.For a routeri,it will receive two types of Interest packets.One refers to those that can be routed to data sources and get Data packets directly,and the other is those that cannot get corresponding Data packets owing to duplicates or Interest aggregation.We assume that routerireceives αrequests per second and the proportion of requests that will be aggregated and lead to the unanticipated data access failure or recognized as duplicates by upstream routers isp.

In NDN,for those requests that are aggregated and lead to the unanticipated data access failure or recognized as duplicates by upstream routers,the corresponding PIT entries will exist in the routers until expiration.Hence,the PIT storage size at routeriis

whereTexis the expiration time of PIT entries,Tdis the average time that routeritakes to get Data packets,andPITavgis the average size of a PIT entry.

In contrast,the PIT storage size at routeriis

if NDN is enhanced with NDRUDAF.With NDRUDAF,for those requests that are aggregated and fail to retrieve data at the first time or judged as duplicates,routeriwill receive NACK packets as responses and refetch as soon as possible if necessary.Tnis the average pending time of PIT entries for such requests at routeri.

Therefore,compared with NDN,the mechanism we proposed can reduce storage by

In general,routers should quickly get a NACK packet,while the expiration time of a PIT entry is always large enough to ensure that requests can arrive at data sources and the corresponding Data packets can return back.Namely,theTnis negligible as compared toTex.Furthermore,as the requests arrival speed αshould be large,the storage space that NDRUDAF saves is significant.And the larger the proportionpis,the more the storage can be saved.

VI.PERFORMANCE EVALUATION

This section presents experimental studies on the performance of the NDN enhanced with NDRUDAF and compares it with that of NDN.

6.1 Experimental setup

We use the open-source ndnSIM[11],a NS-3 based network simulator,to simulate the NDN protocol stack and implement NDRUDAF.We run our simulations on the topology shown in figure 1.We configure the delay over each link to 1msexcept the link between routerR6 and routerR7.The delay of the link betweenR6 andR7 is set to 1.2ms.The bandwidth of each link is set to 100Mbps.The adopted forwarding strategy is flooding so that Interests would be forwarded via different paths.ConsumerC1 sends Interests forprefixandprefix1,and consumerC2 sends Interests forprefixandprefix2.The Interests forprefixare the ones that are going to be aggregated.The requests forprefix1 andprefix2 follow the Zipf-Mandelbrot distribution [12] with parametersq=0.7 ands=0.7.The numbers of distinct content objects underprefix1 andprefix2 are both set to 1000.These requests forprefix1 andprefix2 and the data traffic that they bring back serves as the background traffic.In our simulations,Interests forprefixsent byC1 would arrive at routerR4 earlier than those byC2 due to less hop counts.Then the latter would be aggregated atR4.Furthermore,Interests forprefixsent byC1 routed via the pathR4-R5-R6-R7 will be recognized as duplicate ones atR7,sinceR7 has already received the same requests via the other pathR1-R2-R3-R7.In terms of these duplicate Interests,R7 will simply drop these Interests in NDN,but in the NDN enhanced with NDRUDAF,R7 will send a corresponding NACK Interest with a NACK code “Duplicate”.In addition,R4 would retransmit the aggregated Interests issued byC2 after receiving the NACK Interest in NDRUDAF.In our simulations,the Interests are sent by consumers at a constant rate of 100 Interests per second.The percentage of Interests forprefixsent by each consumer,i.e.,the percentage of aggregated InterestsfromC2,is set to be the same,which is 0%,20%,40%,60%,80%,and 100% in different simulations.For example,when the percentage of aggregated Interests is set to 20%,the consumerC1 produces 20 Interest packets forprefixand 80 Interest packets forprefix1 per second,and the consumerC2 produces 20 Interest packets forprefixand 80 Interest packets forprefix2 per second.The default lifetime of an Interest is configured to 1 second and the lifetime of a retransmitted Interest is doubled.The lifetime of each PIT entry is configured to the lifetime of the Interest that it describes.Each simulation is repeated for 10 runs to get the average results.Each run is set with NS-3 “RngRun” argument as seed to randomize the requests forprefix1 andprefix2 and lasts for 90 seconds.Table 1 shows the setting of primary parameters in our simulations.

Table I.Parameter setting.

Fig.2.Average data access delay versus the percentage of aggregated Interests.

6.2 Experimental results

We evaluate the performance of the NDN enhanced with NDRUDAF from the following three aspects:

·the average data access delay:the average time experienced by a consumer for retrieving the desired content.

·the average PIT entry pending time:the average time that a PIT entry remains in the PIT waiting for the desired Data or a NACK to return.

·the average PIT size:the average number of PIT entries maintained by routers.

1) Impact on Average Data Access Delay:Figure 2 shows the average data access delay versus the percentage of aggregated Interests.It can be seen that the average data access delay in the NDN enhanced with NDRUDAF is much smaller than that in NDN under the considered percentages of aggregated Interests.For example,when the percentage of aggregated Interests is set to 20%,the average data access delay in NDN is 200 times more than that in the NDN enhanced with NDRUDAF.The average data access delay in NDN dramatically increases with the percentages of aggregated Interests,while that in the NDN enhanced with NDRUDAF only slightly increases with the percentages of aggregated Interests.The reason is that in NDN,for each aggregated Interest that fails to bring back the requested Data packet,ConsumerC2 can not realize the data access failure until the PIT entry for the aggregated Interest times out.It resends the Interest which brings back the desired content within another round-trip time to a cache or the producer.Namely,the time for the data access from each aggregated Interest is huge and is the lifetime of the Interest,i.e.,1 second,plus the time for the second data retrieval.Then the average data access delay would dramatically increases with the percentage of aggregated Interests.When the percentage of aggregated Interests increases to 100%,the average data access delay is actually the average time for the data access from each aggregated Interest,which is more than 1000ms.In contrast,in the NDN enhanced with NDRUDAF,for each aggregated Interest that fails to bring back the requested Data packet,its data access delay is the time transmitting the retransmitted Interest from the aggregating node to the data source,which can be a cache2Since the aggregated Interest is immediately retransmitted,it has more chance to get a cache hit as compared to that in NDNor the producer,plus the time delivering the requested content to the consumer,which is relatively much smaller.Then the average data access delay also increases,but slightly,with the percentage of aggregated Interests.

2)Impact on PIT Entry Pending Time:Figure 3 displays the average PIT entry pending time versus the percentage of aggregated Interests.As shown,the average PIT entry pending time in NDN drastically increases with the percentage of aggregated Interests.The reason is that routers which receive duplicate Interests will simply drop the duplicate Interests while the downstream routers are still waiting for the corresponding data.The corresponding PIT entries of downstream routers will exist until their lifetime expires.Therefore,the average PIT entry pending time in NDN is related to both the PIT entry lifetime and the roundtrip time for delivering a desired Data packet.The PIT entry lifetime is far longer than the average round-trip time between consumers and producers.As the percentage of aggregated Interests increases,the average PIT entry pending time gets closer and closer to the PIT entry lifetime which is 1000ms.In contrast,in the NDN enhanced with NDRUDAF,the average PIT entry pending time slightly increases as the percentage of aggregated Interests.The pending time of the PIT entries for the aggregated Interests is related to the round-trip time between consumers and producers plus that between routerR4 and the producer.The pending time of the PIT entries for other Interests should be less than the round-trip time between consumers and producers as these Interests may get cache hits.As the percentage of aggregated Interests increases,the average PIT entry pending time gets closer and closer to the pending time of the PIT entries for the aggregated Interests.

Fig.3.Average PIT entry pending time versus the percentage of aggregated Interests.

3) Impact on PIT Size:Figure 4 illustrates the average PIT size versus the percentage of aggregated Interests.As illustrated,the average PIT size in NDN drastically increases with the percentage of aggregated Interests.The reason is that the PIT entries corresponding to the Interests that cannot get desired Data owing to Interest aggregation and multipath forwarding will exist until these PIT entries expired,and these PIT entries will accumulate quickly at the on-path routers.While in the NDN enhanced with NDRUDAF,the average PIT size only slightly increases with the percentage of aggregated Interests and it is much smaller even when the percentage of aggregated Interests is 0,i.e.,no Interest aggregation.The reason for the former is that the desired Data packets for the aggregated Interests can be brought back quickly and their PIT entries are consumed soon,even though the pending time of such PIT entries is a little greater than that of the PIT entries for the Interests that are not aggregated.The reason for the latter is that the Interests forprefix1 andprefix2 are transmitted via two paths and would be recognized as duplicates by routerR7.The PIT entries for such Interests on the paths via which the Interests were recognized as duplicates would keep pending until they expire in NDN,but they would be instantly purged by NACK packets in the NDN enhanced with NDRUDAF.

Fig.4.Average PIT Size versus the percentage of aggregated Interests.

The simulation results show that the Interest aggregation and multipath forwarding has a enormously negative impact on PIT size in such scenarios and NDRUDAF can greatly improve this situation.

VII.CONCLUSION

Interest aggregation and natural support for multipath forwarding are widely regarded as two significant features that improve the system performance of Named Data Networking (NDN).However,in this paper,we state that the simultaneous adoption of multipath forwarding and Interest aggregation,in certain case,may lead to the data access failure at the consumer whose Interest is aggregated even though the desired Data is available and the Interest is correctly forwarded by the network.To the best of our knowledge,we are the first that identify such data access failure and we further propose NDRUDAF,a NACK based mechanism that enhances the Interest forwarding and then enables detection and fast recovery from such unanticipated data access failure.NDRUDAF relies on the NACK Interest with code “Duplicate” which is sent by the upstream node identifying the duplicate Interest from multiple paths to notify downstream nodes of the duplication.On the receipt of the NACK Interest with code “Duplicate”,the node that aggregates the Interests tries to retrieve the desired Data packet again by forwarding the aggregated Interest upstream.Then the consumer whose Interest is aggregated would receive the desired Data packet soon without knowing that it has ever been involved in a data access failure.We qualitatively analyze the performance of the NDN with our proposed NDRUDAF and compare its performance with that of NDN.Besides,we conduct experimental studies on the performance of the NDN enhanced with NDRUDAF.Our experimental results demonstrate that NDRUDAF indeed improve the system performance in case of the unanticipated data access failure in terms of data access delay and network resource utilization efficiency at routers as compared to the present NDN.

ACKNOWLEDGMENT

This work was supported in part by the National Natural Science Foundation of China (No.61602114),in part by the National Key Research and Development Program of China (2017YFB0801703),in part by the CERNET Innovation Project (NGII20170406),and in part by Jiangsu Provincial Key Laboratory of Network and Information Security (BM2003201).