W. Fang and L. Peterson, "Inter-AS traffic patterns and their implications, " in Proc. IEEE Global Internet Symp., Rio de Janeiro, Brazil, Dec. 1999.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....level measurement capabilities. Figure 1 provides an indication of the accuracy of the method. The method is remarkably accurate for the all but the smallest entries in the traffic matrix. We note that the larger values in the traffic matrix dominate network and traffic engineering applications [14, 15, 13]. The majority of the traffic lies within 23 relative error, and a more than 30 of the matrix elements have negligible error. A more detailed examination of the data will show that the relative errors are largest for the smallest matrix elements, which fortunately do not have large absolute ....

W. Fang and L. Peterson, "Inter-AS traffic patterns and their implications," in Proceedings of IEEE GLOBECOM '99, (Rio de Janeiro, Brazil), pp. 1859--1868, 1999.

....OSPF is mainly used for intra domain traffic routing. It routes traffic on the shortest path based on the advertised link weights. As a result, the link along the shortest path between two nodes may become congested while the links on longer paths may remain idle. Many traffic measurements[43] [44] have observed large variations in link utilization across the network. OSPF allows for Equal Cost Multi Path(ECMP) where traffic is distributed equally among various next hops of the equal cost paths between a source and a destination [45] This is useful in distributing the load to several ....

....In other words, the traffic demand is split into flows at the level of per destination prefix. A typical BGP routing table consists of thousands of destination prefix entries. It will be very complex to work with such a large number of traffic flows. However, many traffic measurements [43] [44] have demonstrated the existence of so called elephant and mice phenomenon. That is, a small number of traffic streams, known as elephants, generate a large portion of total traffic whereas a large number of streams, mice, generate a small portion of total traffic. For example, it has been found ....

[Article contains additional citation context not shown here]

Wenjia Fang and Larry Peterson. Inter-as traffic patterns and their implications. In Proceedings of Global Internet 99, Rio, Brazil, 1999.

....prefixes. For example, traffic destined for the top 1 of the prefixes is responsible for about 20 of the outbound traffic volume. The top 10 of prefixes accounts for approximately 70 of the traffic. These results are consistent with the trends seen in earlier traffic measurement studies [21, 15, 22]. The results are more dramatic when we group prefixes with the same routing choices, as shown by the middle curve in Figure 12. For example, 10 of the 20,000 sets of routing choices contribute more than 80 of the traf 1e 05 0.0001 0.001 0.01 0.1 1 10 100 1000 10000 100000 of Traffic ....

W. Fang and L. Peterson, "Inter-AS traffic patterns and their implications," in Proc. IEEE Global Internet Symposium, December 1999.

....Internet. Then, we analyze the variability of the interdomain flows by considering their activity and the variability of their traffic in volume. I. INTRODUCTION The behavior of Internet traffic has been analysed by researchers almost since the creation of the first computer networks [KN74] FP99] CBP93] TMW97] Many studies have tried to better understand the microscopic or packet level behavior of the Internet. In this case, the analysis is usually performed on the basis of a capture of all the packets that flow through a given link for some period of time. This type of analysis has ....

....http: www.infonet.fundp.ac.be. E mail: fsuhlig,obonaventureg info.fundp.ac.be. This work was supported by the European Commission within the IST ATRIUM project. the basis of packet level traces and SNMP statistics from backbone routers. They confirmed the findings of [KN74] More recently, FP99] analyzed several one hour packet level traces from universities and a commercial backbone to evaluate the impact of aggregating flows at the Autonomous System level. A similar analysis was used in [PHS00] to evaluate the performance of an interdomain reservation scheme. FFF99] has also ....

[Article contains additional citation context not shown here]

W. Fang and L. Peterson. Inter-AS traffic patterns and their implications. In IEEE Global Internet Symposium, December 1999.

.... the value of f(x ) in the first n samples as the threshold value y r (line 4) and any future sample with a smaller function value than y r is considered to belong to AD (r) In later exploration, a new x is obtained every n samples and y r is updated with the average of these x (lines [21 23]) Note that this calculation of y r is not intended to be an accurate estimation of the threshold for AD (r) instead it only function as the adjustment for the balance between exploration and exploitation. In other words, it is to ensure that on the average the exploration process will Algorithm ....

....of Service(QoS) constraints. As a result, the routing generated by these algorithms tend to generate a highly uneven mapping of traffic. Some links may get very congested since most of traffic go through it and the other may be underutilized. This has been observed in many traffic measurements[20, 21]. Suppose the offered traffic load in a network is defined by a demand matrix, where the row index represents the source, the column index the destination and the element of the matrix the offered load from the source to the destination. Such demand matrix can be obtained through network ....

Wenjia Fang and Larry Peterson. Inter-as traffic patterns and their implications. In Proceedings of Global Internet 99, Rio, Brazil, 1999.

....protocols are often called topology driven. The routing generated by such algorithms tends to result in a highly uneven mapping of traffic. Some links may get very congested while the others may be consistently under utilized. This phenomenon has been confirmed by many traffic measurements[2] [3] where a large variation in link utilizations is observed across the network. Traffic Engineering (TE) tries to eliminate this situation by adapting the network routing according to the prevailing traffic conditions. Traffic engineering deals with mapping traffic to physical network topology in ....

....base (MIBs) packet sampling etc. to obtain the traffic demand statistics [12] A typical BGP routing table consists of thousands of entries for various destination prefixes. It will be very complex and undesirable to work with such a large number of traffic flows. Many measurements [2] [3] have demonstrated that the Internet traffic exhibits the so called elephant and mice phenomenon. A small number of traffic streams, known as elephants, generate a large portion of total traffic whereas a large number of steams, the mice, generate a small portion of total traffic. Another ....

[Article contains additional citation context not shown here]

Wenjia Fang and Larry Peterson. Inter-as traffic patterns and their implications. In Proceedings of Global Internet 99, Rio, Brazil, 1999.

....level measurement capabilities. Figure 1 provides an indication of the accuracy of the method. The method is remarkably accurate for the all but the smallest entries in the traffic matrix. We note that the larger values in the traffic matrix dominate network and traffic engineering applications [14, 15, 13]. The majority of the traffic lies within 23 relative error, and a more than 30 of the matrix elements have negligible error. A more detailed examination of the data will show that the relative errors are largest for the smallest matrix elements, which fortunately do not have large absolute ....

W. Fang and L. Peterson, "Inter-AS traffic patterns and their implications," in Proceedings of IEEE GLOBECOM '99, (Rio de Janeiro, Brazil), pp. 1859--1868, 1999.

....packets. Within each group, flows arrive as a Poisson process with rate #.Wevary# to study both uncongested and congested scenarios. We use the ns 2 built in routines to generate web sessions consisting of a single object each. This is what we Notice that in accordance with the usual practice [8], 12] 13] packets are said to belong to the same flow if they have the same source and destination IP address, and source and destination port number. A flow is on if its packets arrive more frequently than a timeout of some seconds. This timeout is usually set to something less than 60 ....

W. Fang and L. Peterson. Inter-as traffic patterns and their implications. In Proceedings of the 4th Global Internet Symposium, December 1999.

....the hardware currently available, due to the computational overhead of maintaining per flow information for all flows, but is useful to assess the viability of our design. To address scalability issues, we note that in practice, only a small number of flows contribute to the majority of traffic [2, 12, 13]. We conjecture that tracking and marking only these heavy hitters is sufficient for avoiding packet drops. Based on this idea of filtering flows, we present a heuristic approximation of the reference algorithm that we expect to be computationally efficient enough to be deployed in edge routers, ....

....to reduce the number of tracked TCP flows, and employs linear interpolation to reduce the computational complexity of the projection algorithm. Our goal is to design a heuristic algorithm that is deployable in an edge or an access router. 3. 1 Flow Filtering As observed in measurement studies [2, 12, 13], only a small percentage of flows ( heavy hitters ) accounts for a large percentage of traffic. These heavy hitters transmit at a high data rate due to (1) a large congestion window, and (2) a relatively small round trip time. From the description of the reference algorithm in Section 2, these ....

W. Fang and L. Peterson. Inter-AS traffic patterns and their implications. In Proceedings of IEEE GLOBECOM '99, pages 1859--1868, Rio de Janeiro, Brazil, December 1999.

....arrivals at the router, and avoid such impending conflicts. Unfortunately, the proposed method requires to maintain per flow information, which defeats our scalability criteria. However, it has been shown that a very small number of flows, or heavy hitters , contribute to the majority of traffic [13]. Thus, sampling techniques such as multistage filters can be used to filter out heavy hitters , and drastically reduce the amount of state information to be maintained [14] Note that the proposed technique resembles proactive marking dropping algorithms such as RED [15] 11 0 100 Class 1 ....

W. Fang and L. Peterson, "Inter-AS traffic patterns and their implications," in Proc. IEEE GLOBECOM '99, Rio de Janeiro, Brazil, Dec. 1999, pp. 1859--1868.

....is. A high bandwidth link may see more unique destination addresses than a low bandwidth link but the increase in addresses seen may not be proportional to the increase in bandwidth. e.g. It is known that for inter AS traffic a small percentage of end host flows contribute to a large percentage [8] of the traffic. The packet size distribution depends on the number of requests and the file sizes requested. With web traffic the packet sizes usually vary from 40 bytes (connection setup packets) to 1500 bytes (the path maximum transmission unit (MTU) is the minimum of the maximum ....

Fang, W. and Peterson, L., "Inter-AS traffic patterns and their implications," Proceedings of Global Telecommunications Conference, Vol. 3, pp. 1859-1868, Rio de Janeiro, Brazil (Dec 1999).

.... there are many studies on the topology of the Internet (see [11] and the references therein) or the evolution of the BGP routing tables (see [5] and the references therein) as well as many studies on the packet level characteristics of the traffic (see [12] among many other papers) few papers [6], 7] 13] analyze together the traffic and its topological distribution. In the framework of a detailed analysis of interdomain traffic, we have collected several traces of all the traffic received or sent through the border routers of three stub ISPs. Due to practical reasons, it was ....

W. Fang and L. Peterson. Inter-as traffic patterns and their implications. In IEEE Global Internet Symposium, December 1999.

....that a trial would be still possible on the existing exchange point. Appendix B AS by Connectivity In the following table we give the best connected AS numbers, outdegree and network names that were used in the evaluation of this work. These ASes correspond roughly to the class 1 and 2 networks [Fang and Peterson, 1999]. Data is from [moat, 1999] and names were translated by whois.radb.net. 701 1186 Alternet 3561 692 Cable Wireless (CW) 1239 573 SprintLink Backbone 1 301 GTE Internetworking 7018 271 AT T WorldNet Service Backbone 2548 248 DIGEX AS 2914 244 6453 149 Teleglobe Canada Inc. 293 138 ....

FANG, W. AND PETERSON, L. Inter-AS Traffic Patterns and Their Implications. In Global Internet 99, Rio de Janeiro, Brazil. December 1999.

....stable BGP routes [7] A relatively small number of prefixes are responsible for the bulk of the BGP update messages. Several recent studies have made a similar observation about traffic volumes a small fraction of the destination prefixes are responsible for the majority of Internet traffic [8 10]. This paper asks a simple, yet important, question about the relationship between BGP updates and traffic volumes: Do the small fraction of popular prefixes have relatively stable BGP routes On the one hand, popular destinations would presumably have reliable and well managed connections to the ....

....received little or no traffic. The graph ranks the prefixes by the volume of traffic and plots the proportion of the traffic from the highest ranked prefix to the lowest. For example, the most popular prefix in each direction contributes more than 1 of the traffic, consistent with other studies [8 10]. B. BGP Stability and Prefix Popularity The prefixes responsible for most of the events do not receive much of the traffic, as shown by Figure 5(a) To construct this graph, we ranked the prefixes by the number of update events (using a 45 second timeout) and plotted the cumulative distribution ....

W. Fang and L. Peterson, "Inter-AS traffic patterns and their implications, " in Proc. IEEE Global Internet, December 1999.

....forms the basis for building components of POPto POP level traffic matrices, which are key to studying a variety of traffic engineering and routing issues. We investigate how much can be said about the traffix matrix just from the data collected at a single POP. It has been observed ( 2] 3] [4]) that obtaining information about traffic patterns in both time and space is critical for most traffic engineering functions. Traffic engineering typically operates on long time scales such as minutes, hours, weeks or longer. Examples of traffic engineering functions include dimensioning, ....

....of these aggregate streams, called elephants, generate a large fraction of the total traffic, while a large number of these streams, called mice, generate a small fraction of the total traffic. The elephants and mice phenomenon has been observed before in Internet traffic at the inter AS level [4], at the level of multipoint demands from one router node to a set of router nodes [3] and in the Internet as it was many years ago [8] Here we demonstrate this phenomenon at the granularity level of specific prefixes. We also demonstrate the stability of these aggregates throughout the day. The ....

[Article contains additional citation context not shown here]

W. Fang and L. Peterson, "Inter-AS Traffic Patterns and Their Implications," Proceedings of Global Internet, December 1999.

....The influence of the variations of the BGP routing table on the interdomain traffic is clearly outside the scope of this chapter. Using a single BGP table for each ISP is an approximation but since we rely on the BGP table of the studied ISPs our analysis is more precise 10 than other studies [FP99, PHS00] that relied on a BGP routing table collected at a different place and time than the packet traces studied in these papers. The routing table of the dialup ISP contained 102345 active prefixes, covering about 26 of the total IPv4 address space. This coverage of the total IPv4 address ....

W. Fang and L. Peterson. Inter-AS traffic patterns and their implications. In IEEE Global Internet Symposium, December 1999.

....the Internet. Then, we analyze the variability of the interdomain flows by considering their activity and the variability of their traffic in volume. 1 Introduction The behavior of Internet traffic has been analysed by researchers almost since the creation of the first computer networks [KN74, FP99, CBP93, TMW97] Many studies have tried to better understand the microscopic or packet level behavior of the Internet. In this case, the analysis is usually performed on the basis of a capture of all the packets that flow through a given link for some period of time. This type of analysis has ....

....from this paper was that a few sites were responsible for a larger portion of the traffic. Similar studies were conducted in 1993 on the NSFNet backbone [CBP93] on the basis of packet level traces and SNMP statistics from backbone routers. They confirmed the findings of [KN74] More recently, FP99] analyzed several one hour packet level traces from universities and a commercial backbone to evaluate the impact of aggregating flows at the Autonomous System level. A similar analysis was used in [PHS00] to evaluate the performance of an interdomain reservation scheme. FFF99] This work was ....

[Article contains additional citation context not shown here]

W. Fang and L. Peterson. Inter-AS traffic patterns and their implications. In IEEE Global Internet Symposium, December 1999.

.... In addition, BELNET was directly connected to the Belgian interconnection point (BNIX) and with a Mbps link to the Dutch national interconnection point (AMS IX) Many researchers have analysed the behavior of Internet traffic based on measurements gathered in operational networks [KN74, FP99, CBP93, TMW97] Often, the analysis focuses on the packet captured on a single or a few different links [FP99, TMW97] for a relatively short period of time. These packet traces are very precise but require a large storage space. Other studies have recently relied on less precise information like ....

.... link to the Dutch national interconnection point (AMS IX) Many researchers have analysed the behavior of Internet traffic based on measurements gathered in operational networks [KN74, FP99, CBP93, TMW97] Often, the analysis focuses on the packet captured on a single or a few different links [FP99, TMW97] for a relatively short period of time. These packet traces are very precise but require a large storage space. Other studies have recently relied on less precise information like SNMP statistics [CBP93] For this paper, we rely on a trace that differs by several aspects from the traffic ....

[Article contains additional citation context not shown here]

W. Fang and L. Peterson. Inter-AS traffic patterns and their implications. In IEEE Global Internet Symposium, December 1999.

....of traffic over fewer heavy hitters in outbound versus inbound traffic. Similar trends have been seen in earlier studies that consider the load on individual links or servers. For example, link level traces show that the distribution of traffic at the prefix and AS level follows Zipf s law [19]. Studies of the World Wide Web have shown that a small fraction of the requests, resources, and servers are responsible for the bulk of the traffic [20, 21] The small number of heavy hitters has important implications for traffic engineering. On the positive side, since the leading heavy hitters ....

W. Fang and L. Peterson, "Inter-AS traffic patterns and their implications," in Proc. IEEE Global Internet Symposium, December 1999.

....pricing methods and business strategies are used to calculate prices for new offers 3 . 2.1 SLA Trading Optimization Objectives SLA trading is performed by SLA traders situated in the ISP s DS domain. For the time being we assume traders to be cen2 Jumbo flows are defined between two ASes [10]. 3 For the sake of simplicity we assume global currency. tralized for each AS. SLA traders make local decisions about what services are provided to which peers. Such decisions are made spontaneously or they are the reaction to an external event. Initially, SLA traders at ISP i may offer ....

....to model inter AS flows with high burstiness across time and aggregation scales. The spatial inter AS traffic distribution is suspected to be highly non uniform (hot spots) 12] This is supported by a recent study on the distribution of jumbo flows (source AS to destination AS aggregate flow) [10]. Such a jumbo flow model has been used for simulations network that shows such a highly non uniform locality of flows. The model is defined by the parameters given in Table 4. The main output parameter is utilization of the network. We load a network configuration until all SLA traders in the ....

[Article contains additional citation context not shown here]

W. Fang and L. Peterson. Inter-AS Traffic Patterns and Their Implications. In Global Internet 99, Rio de Janeiro, Brazil, December 1999.

No context found.

W. Fang and L. Peterson, "Inter-AS traffic patterns and their implications, " in Proc. IEEE Global Internet Symp., Rio de Janeiro, Brazil, Dec. 1999.

No context found.

FANG, W., AND PETERSON, L. Inter-AS Traffic Patterns and Their Implications. In Proc. of IEEE Global Internet Symposium (1999).

No context found.

W. Fang and L. Peterson. Inter-AS traffic patterns and their implications. In Proc. IEEE Global Internet Symposium, Rio, Brazil, December 1999.

No context found.

W. Fang. and L. Peterson, Inter-AS traffic patterns and their implications. Proceedings of IEEE Global Telecommunications Conference 1999.

No context found.

W. Fang and L. Peterson, "Inter-AS traffic patterns and their implications," in IEEE Global Internet Symposium, December 1999.

No context found.

W. Fang and L. Peterson. Inter-AS traffic patterns and their implications. In IEEE Global Internet Symposium, December 1999.

No context found.

W. Fang and L. Peterson, "Inter-AS traffic patterns and their implications," in IEEE Global Internet Symposium, December 1999.

No context found.

W. Fang and L. Peterson. Inter-AS Traffic Patterns and Their Implications. IEEE Globecom, December 1999. Brazil.

No context found.

W. Fang and L. Peterson. Inter-AS traffic patterns and their implications. In IEEE Global Internet Symposium, December 1999.

No context found.

Fang, W, and Peterson, L., "Inter-as Traffic Patterns and their Implications ", Proceedings of IEEE GLOBECOM 1999.

No context found.

W. Fang and L. Peterson. Inter-AS Traffic Patterns and Their Implications. IEEE Globecom, December 1999. Brazil.

No context found.

W. Fang and L. Peterson, "Inter-AS traffic patterns and their implications," in Proc. IEEE Global Internet Symposium, December 1999.

No context found.

W. Fang, L. Peterson, Inter-AS traffic patterns and their implications, in: Proceedings of IEEE GLOBECOM '99, Vol. 3, Rio de Janeiro, Brazil, 1999, pp. 1859--1868.

No context found.

W. Fang and L. Peterson, "Inter-AS traffic patterns and their implications," in Proc. IEEE Global Internet Symposium, December 1999.

No context found.

W. Fang and L. Peterson. Inter-AS traffic patterns and their implications. In Proceedings of IEEE Globecom: Global Internet, Rio de Janeiro, Brazil, December 1999.

No context found.

W. Fang and L. Peterson. Inter-AS traffic patterns and their implications. In IEEE Global Internet Symposium, December 1999.

No context found.

W. Fang and L. Peterson. Inter-AS Traffic Patterns and Their Implications. IEEE Globecom, December 1999. Brazil.

No context found.

W. Fang and L. Peterson. Inter-AS traffic patterns and their implications. In IEEE Global Internet Symposium, December 1999.

No context found.

W. Fang and L. Peterson. Inter-AS traffic patterns and their implications. In IEEE Global Internet Symposium, December 1999.

No context found.

W. Fang and L. Peterson, "Inter-AS Traffic Patterns and Their Implications," IEEE Globecom, December 1999, Brazil.

No context found.

W. Fang and L. Peterson. Inter-AS traffic patterns and their implications. In Proceedings of IEEE GLOBECOM '99, volume 3, pages 1859--1868, Rio de Janeiro, Brazil, December 1999.

No context found.

W. Fang and L. Peterson. Inter-as traffic patterns and their implications. In IEEE Global Internet Symposium, December 1999.

No context found.

W. Fang and L. Peterson. Inter-AS traffic patterns and their implications. In IEEE Global Internet Symposium, December 1999.

No context found.

W. Fang and L. Peterson, "Inter-AS Traffic Patterns and Their Implications," IEEE Global Internet Symp., Dec. 1999.

No context found.

W. Fang and L. Peterson. Inter-AS traffic patterns and their implications. In IEEE Global Internet Symposium, December 1999.

No context found.

W. Fang and L. Peterson. Inter-AS traffic patterns and their implications. In IEEE Global Internet Symposium, December 1999.

No context found.

W. Fang and L. Peterson, "Inter-AS traffic patterns and their implications," in Proc. IEEE Global Internet Symposium, December 1999.

No context found.

Wenjia Fang and Larry Peterson. Inter-as traffic patterns and their implications. In Proceedings of Global Internet 99, Rio, Brazil, 1999.

No context found.

W. Fang and L. Peterson, "Inter-AS traffic patterns and their implications," in Proc. IEEE Global Internet Symposium, December 1999.

No context found.

W. Fang and L. Peterson, "Inter-AS traffic patterns and their implications," in Proceedings of IEEE GLOBECOM '99, pp. 1859--1868, (Rio de Janeiro, Brazil), Dec. 1999.

First 50 documents  Next 50

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC