Claffy, K., Braun, H. W., and Polyzosi, G., "Traffic characteristics of the T1 NSFNET backbone," Proceedings of IEEE Infocom, pp. 885-892, San Francisco (Aug 1993).  Home/Search   Document Not in Database   Summary   Related Articles   Check

....we expect that the packet size distribution in the generated trace will more closely resemble the real trace. 5.3 Destination addresses Figure 5 shows the CDF for the number of unique destination addresses seen in an interval of 1 second. Studies of traffic locality in the NSFNET backbone [5] indicate that a large percentage of the traffic is destined for a small percentage of hosts. Other studies [4] indicate that the number of host pairs increases as the square root of the bit rate. These effects should be considered when evaluating the generated traces. The diversity of the ....

Claffy, K., Braun, H. W., and Polyzosi, G., "Traffic characteristics of the T1 NSFNET backbone," Proceedings of IEEE Infocom, pp. 885-892, San Francisco (Aug 1993).

....a paradigm that assumes network link symmetry. Many fixed networks and satellite or wireless networks exhibit asymmetry in capacity or load. In addition, traditional multicast routing algorithms utilize reverse path routing which in asymmetric networks may lead to poor routes. Several studies [10, 3] have been made that confirm the existence of network asymmetry in the fixed networks that make up the intemet There are two basic approaches to multicast tree construction. The first is a shared multicast tree [1] and the other is a source rooted tree [13, 9] The shared tree approach uses a ....

....interconnected with a network generated using RG1 and RG2. Topologies varied from 10 100 nodes distributed over 1 10 clusters. Link cost was defined as a uniform random variable between 1 100 for intra cluster links and 1 1000 for inter cluster links. Average node degree was kept in the interval [3,5] for each topology. The simulations also allowed several sessions to be built on top of each other so that load balancing over several sessions could be studied. The shared trees were constructed as per the protocols presented earlier The shared trees were constructed around centers selected using ....

K. Claffy, G. Polyzos, and H. W. Braun, "Traffic Characteristics of the T1 NSFNET Backbone," Proc. oflEEE INFOCOM '93, March 1993.

....area network traffic and has reported on packet inter arrival and size distributions. Shoch and Hupp s landmark paper [37] is perhaps one of the earliest papers on traffic measurements over a Local Area Network. Traytic Characteristics of the T1 NSFNet BackBone: A recent study by Claffy et al. [4] has reported on the traffic characteristics over the NSFNet backbone. The study gives long term growth of traffic volume, packet size statistics for different protocols, distribution of median delays across different backbone nodes, link utilization statistics, and link error and nodal downtime ....

Claffy, K.C., G.C. Polyzos and H.W. Braun, "Traffic characteristics of the T1 NSFNet backbone," ?roc. of the IEEE Infocorn, to appear.

....links incident to node n i , i.e. W i = # d i j=1 w ij . Then, instead of selecting a node of minimal degree, the generalized algorithm will select a node n i with minimum weight W i (excluding the special node, if there is such one) For instance, in Fig. 10 (inspired by the NSFNET topology [8]) node NE with weight W NE =4is first selected. Next, node CO with weight W CO =4is selected (remind that the links incident to node NE have been deleted in the previous step) The procedure continues until all the nodes have been considered and deleted. Using an approach similar to the ....

K. Claffy, G. Polyzos, and H.-W. Brauns, "Traffic Characteristics of the T1 NSFNET Backbone," in the proceedings of INFOCOM '93, pp. 885-892, San Francisco, CA.

....retransmission schemes may require a different retransmission timeout (RTO) estimator and leads us to believe that research in this area should be extended. We should further mention that the Internet has been extensively studied by various researchers using ICMP ping and traceroute packets [1] [9], 16] 17] 20] UDP echo packets [6] 7] and multicast backbone (MBone) audio packets [24] With the exception of the last one, similar observations apply to these studies neither the setup, nor the type of probe traffic represented realistic real time streaming scenarios. In addition, ....

K. Claffy, G.C. Polyzos, and H-W. Braun, "Traffic Characteristics of the T1 NSFNET Backbone," IEEE INFOCOM, 1993.

....consultation and diagnosis over geographically distributed regions using multimedia communications, distributed databases, and network protocols. The national Internet and its supporting backbone network have link bandwidths of T1 (1. 5 Mbps) and T3 (45 Mbps) respectively, across the United States [11]. 3 Local DBASs Local DBMS DBAS Local PACS Remote WSs Remote DBASs Remote DBMS DBAS Remote DBASs Remote WSs Remote PACS Remote PACS Local WSs Imaging Equipments Imaging Equipments Imaging Equipments Remote WSs SLIP Server telephone line Rural and Global PACS Backbone Network ....

G. Polyzos, K. Claffy, and H. Braun, "Traffic Characteristics of the T1 NSFNET Backbone," Working Conference Paper, May 1993.

.... are Kleinrock s study of the ARPANET s behavior on time scales of hours to days [Kl76] the series of ping experiments conducted by Mills to evaluate the effectiveness of the TCP retransmission timeout algorithm [Mi83] Claffy et al. s study characterizing traffic on the T1 NSFNET backbone [CPB93b]; and Chinoy s study of the dynamics of routing information within the NSFNET backbone [Ch93] While these studies can convincingly characterize the full range of behavior one might expect to observe from the network, they become impractical as the 2 network grows in size. 1 Site studies ....

K. Claffy, G. Polyzos, and H-W. Braun, "Traffic Characteristics of the T1 NSFNET Backbone," Proceedings of INFOCOM ' 93, San Francisco, March, 1993.

....Lag ACF 0 10 20 30 40 Series : Generated Figure 6: Section of corresponding generated sequence and its ACF. FDDI, Ethernet, NSFNET External Interfaces and popular TCP protocols. 3 Related Work Properties of network traffic have been the subject of considerable recent interest, see for example [8, 7, 9, 18, 22, 29, 30]. Our current approach is to model differenced series as ARMA processes. Differencing makes a non stationary series stationary when there is a drift in the data. However, not all classes of non stationarities admit a differencing strategy. Some example datasets are given in [4] These show quite a ....

.... we believe that they ought to be considered inclusively in modeling repertoire (based on the results obtained from the two classes on the same datasets) In other related work on Internet traffic, ffl Claffy et al. studied flow parameterization and traffic characteristics of the T1 NSFNET backbone [7, 9]. ffl Danzig et al. [10] developed a traffic library for TCP conversations based on empirical burst distributions they studied in [8] ffl Klivansky et al. studied the impact of different factors on potential long range dependence in aggregate TCP traffic over NSFNET core switches, see [18] ffl ....

Claffy, K. C., G. C. Polyzos and H. W. Braun, "Traffic characteristics of the T1 NSFNet backbone," Proc. IEEE Infocom, San Francisco, 1993.

....planning, and traffic predictions. Periodic usage updates are useful and required by users. A number of recent studies have characterized IP, TCP, and application traffic on the Internet. Measurement studies of IP packet statistics have documented the growth in traffic volume [Paxson 93a, Claffy 93, Rutkowski 94] and the growth of Internet hosts and domains [Lottor 92, Lottor 94, Merit 94, Lynch 93] Some studies focused on end to end behaviour on the Internet [Mills 83, Mukherjee 92, Sanghi 93, Claffy 93] while others focused on the study of wide area traffic [Heimlich 90, Asaba 92, ....

.... of IP packet statistics have documented the growth in traffic volume [Paxson 93a, Claffy 93, Rutkowski 94] and the growth of Internet hosts and domains [Lottor 92, Lottor 94, Merit 94, Lynch 93] Some studies focused on end to end behaviour on the Internet [Mills 83, Mukherjee 92, Sanghi 93, Claffy 93] while others focused on the study of wide area traffic [Heimlich 90, Asaba 92, Wakeman 93, Paxson 91, Caceres 91, Schmidt 92, Danzig 92b, Maffeis 93] 1.1 Thesis Objectives Since the Internet continues to grow rapidly in size as well as in services, users and service providers at the ....

[Article contains additional citation context not shown here]

K. C. Claffy, H. -W. Braun, and G. C. Polyzos, "Traffic Characteristics of the T1 NSFNET Backbone", Proceedings of IEEE INFOCOM '93, Vol. 2, San Francisco, California, July 1993, pp. 885-892.

....been designed to exploit the bimodal pattern of network traffic. There have been many studies of typical TCP IP Internet traffic, and a common trait is the bimodal distribution of packet size. Essentially the traffic consists of either big packets or small packets with little traffic in between [6, 7]. This traffic characteristic is also true with the distributed shared memory environment we are studying [8] The traffic, generated by a memory coherence protocol needed to achieve DSM, has two major forms: memory consistency control packets (such as memory block requests, invalidations, ....

K. Claffy, G. Polyzos, and H. Braun, "Traffic characteristics of the T1 NSFNET backbone," in IEEE INFOCOM'93, pp. 885--892, Mar. 1993.

....on the asymmetry of the network. Our notions of asymmetry are formally defined in section 2.5 and are based on the relative edge costs. The fact that the absence of asymmetry in relative edge costs (as is the case with the (undirected) SMT 1 Studies on network traffic on backbones (e.g. [17]) have revealed marked asymmetry in link utilizations. Moreover, if the link is wireless its bandwidth quality is itself asymmetric due to differences in noise, transmission powers, mobility, etc. of the endpoints. problem) permits twice optimum solutions suggests that asymmetry is crucial to the ....

K. Claffy, G. Polyzos and H.W. Braun, "Traffic Characteristics of the T1 NSFNET Backbone," Proc. IEEE Infocom, Mar. 1993.

....at the T1 NSFNET backbone In 1992, the NSFNET was using T1 (1.544 MBps) links for the backbone. Each backbone node, called NSS, consisted of multiple IBM PC RT processors connected by a token ring, one of which was dedicated to statistics collection by processing every packet on the token ring [8]. Packet Switching Processor Packet Switching Processor Packet Switching Processor Packet Switching Processor Routing Control Statistics Collection Processor Interprocessor Communication Facility (Exterior) Centralized statistics collection was performed with SNMP every 15 minutes. The target of ....

Claffy, Polyzos, Braun, "Traffic Characteristics of the T1 NSFNET Backbone "

....to exploit the bimodalpattern of network traffic. There have been many studies of typical Internet traffic, with TCP IP, and a common trait is the bimodal distribution of packet size: essentially all of the traffic consists of either big packets or small packets with little traffic in between [13, 14]. Batched applications, such as ftp, tend to use large data packets to get high throughput while interactive applications, like telnet, tend to use small packets for low latency. This characteristic is also true with distributed shared memory computer systems. The traffic, generated by a memory ....

K. Claffy, G. Polyzos, and H. Braun, "Traffic characteristics of the T1 NSFNET backbone," in IEEE INFOCOM'93, pp. 885-- 892, Mar. 1993.

....time in ms. Eastern Standard Time Eastern Standard Time Figure 3. Maximum and Median Transmission Delays on the Internet the east coast for packets to New York and Nova Scotia, but much less for ATT Bell Labs (in New Jersey) 17 The time of day variation is also evident in Figure 5(borrowed from Claffy, Polyzos, and Braun (1992)) 18 In Figure 4 we measure delay variation by the standard deviation of delays by time of day for each destination. Delays to Nova Scotia, Canada were extraordinarily variable, yet delays to Oslo were no more variable than in transmission to New Jersey (ATT) Variability in delay fluctuates ....

....Canada Univ of Oslo, Norway 0 200 400 600 800 1000 4 am 8 am 12 pm 4pm Standard Deviation of Delay by Time of Day Roundtrip time in ms. Eastern Standard Time. Figure 4. Variability in Internet Transmission Delays Figure 5. Utilization of Most Heavily Used Link in Each Fifteen Minute Interval (Claffy et al. 1992)) obvious pattern. How much delay is involved, and who is inconvenienced As seen in Figure 3, during our experiment we never experienced delays of more than 1 second in round trip time except to the site in Nova Scotia. Is that too trivial a delay to be concerned about Probably yes , for ....

Claffy, K. C., Polyzos, G. C., and Braun, H.-W. (1992). Traffic characteristics of the T1 NSFNET backbone. Tech. rep. CS92-252, UCSD. Available via Merit gopher in Introducing the Internet directory.

....costs. These asymmetric costs might exist because of unidirectional physical links such as satellite or microwave links or they might be associated with traffic characteristics such as congestion. An example of the latter is the asymmetric traffic loading observed by Claffy on NFSNET T1 links [25]. This chapter presents and compares heuristics suitable for building multicast trees for point to point networks with asymmetric link costs. Many of the heuristics already reviewed in Chapter 2 show promise since they produce good trees and can easily be modified to solve directed networks. In ....

G. Polyzos K. Claffy and H. Braun. "Traffic characteristics of the T1 NSFNET backbone, " in Proc. IEEE INFOCOM, San Francisco, CA, Mar. 1993, pp. 885--892.

....an overall growth rate of about 120 year, though over the most recent eight months the rate has climbed to about 190 year. Unfortunately, the number of different hosts making connections over the backbone is not available. The NSFnet T1 backbone traffic is further studied by Claffy et al. [CPB93], who found that the number of bytes traversing the T1 backbone grew quadratically between June, 1988, and June, 1992, though this growth trend is at least partially influenced by the traffic switching over to the T3 backbone later in the study period. ffl Statistics for USENET network news ....

....the other connections. We only attempted to identify the most popular. In comparison, Rutkowski93] reports 1,066 different identifiable services present on the NSFnet backbone during May, 1993. We also note that the rapid growth of such connections ( 90 year) is in keeping with the finding in [CPB93] that the other protocol traffic on the NSFnet T1 backbone steadily increased. Wenow turn to the number of data bytes (sent in both direc6 Growth in Bytes Day Year 1991 1992 1993 1994 total ftpdata nntp X11 shell www smtp Figure 6: Bytes Day for Largest Protocols tions, and excluding TCP IP ....

[Article contains additional citation context not shown here]

K. Claffy, G. Polyzos, and H.W. Braun, "Traffic Characteristics of the T1 NSFNET Backbone", Proceedings of INFOCOM '93, San Francisco, March, 1993.

....and more generally in the Internet. They have examined this behavior over different time scales. Merit Network Inc. publishes monthly statistics of packet delay between the nodes of the NSFNET. These statistics are obtained from measurements performed at 15 minute intervals. They are used in [5] to examine the distribution of median delay between nodes of the NSFNET. Unfortunately, the Merit statistics are based on measurements performed between the exterior interfaces of the backbone nodes. Thus, they might not accurately characterize the end to end delay over paths which span a ....

K. Claffy, G. Polyzos, H-W. Braun, "Traffic characteristics of the T1 NSFNET backbone", Proc. IEEE Infocom '93, San Fransisco, CA, pp. 885-892, April 1993.

....time, the number of networks connected to the NSFnet has risen exponentially from 2,501 to 17,979, a growth rate of 112 year. Unfortunately, the number of different hosts making connections over the backbone is not available. The NSFnet T1 backbone traffic is further studied by Claffy et al. [CPB93], who find that the number of bytes traversing the T1 backbone grew quadratically between June, 1988, and June, 1992, though this growth trend is at least partially influenced by the traffic switching over to the T3 backbone later in the study period. ffl Statistics for USENET network news traffic ....

....other connections. We only attempted to identify the most popular. In comparison, Rutkowski93] reports 1,066 different identifiable services present on the NSFnet backbone during May, 1993. We also note that the fairly rapid growth of such connections (90 year) is in keeping with the finding in [CPB93] that the other protocol traffic on the NSFnet T1 backbone steadily increased. Wenow turn to the number of data bytes (sent in both directions, and excluding TCP IP headers) per day due to different protocols. Figure 6 plots log 2 of the bytes per day on the Y axis versus the year on the ....

[Article contains additional citation context not shown here]

K. Claffy, G. Polyzos, and H.W. Braun, "Traffic Characteristics of the T1 NSFNET Backbone", Proceedings of INFOCOM '93, San Francisco, March, 1993.

....and digitized at 256 Hz. The y axis range is 195 to 280 microvolts. From [5] data courtesy of EEG Systems Laboratory, San Francisco, CA. B) Computer network activity between two nodes of the NSFNET for 1 week. The y axis ranges from 0 to 80 percent utilization of the link capacity. From [10]) Figure Figure Figure Figure Figure Figure 6: C) HumanEEG as a function of development, from infancy to childhood, illustrating the increasing breadth of frequencies with age. Adapted from [39] See next page. 250ms. Currently the connection between Japan and Germany is via the US and ....

K. Claffy, H.-W. Braun, and G. Polyzos, Jan 1993, Traffic Characteristics of the T1 NSFNET Backbone, SDSC Report GA-A21019, UCSD Report CS92-237, Proceedings of INFOCOM'93

....in both issues argues for replica servers whose establishment is controlled by traffic information collected by proxies. In combination with a simple, application level naming service it also permits to transparently redirect clients to appropriate replica servers [BMS96] Research results as [CPB93] confirming the existence of high traffic favoritism also make the point that the latter approach might be best suited to distribute the most often requested documents farthest into the net. As far as simulations of caching behavior in the WWW are concerned, several in part contradictory ....

K.C. Claffy, G.C. Polyzos, H.-W. Braun: "Traffic Characteristics of the T1 NSFNET Backbone"; Proceedings of SIGCOMM'93, Sep. 1993.

....0 50 100 150 200 250 4 am 8 am 12 pm 4pm Median Delay by Time of Day Roundtrip time in ms. Eastern Standard Time Eastern Standard Time Figure 3. Maximum and Median Transmission Delays on the Internet (in New Jersey) 20 The time of day variation is also evident in Figure 5, borrowed from (Claffy, Polyzos, and Braun (1992)) 21 Figure 4 shows the standard deviation of delays by time of day for each destination. The delays to Canada are extraordinarily variable, yet the delays to Oslo have no more variability than does transmission to New Jersey (ATT) Variability in delays itself fluctuates widely across times of ....

....Canada Univ of Oslo, Norway 0 200 400 600 800 1000 4 am 8 am 12 pm 4pm Standard Deviation of Delay by Time of Day Roundtrip time in ms. Eastern Standard Time. Figure 4. Variability in Internet Transmission Delays Figure 5. Utilization of Most Heavily Used Link in Each Fifteen Minute Interval (Claffy et al. 1992)) technology is that of network control, which entails congestion control, routing control, and bandwidth access and allocation. We expect that if access to Internet bandwidth continues to be provided at a zero cost there will inevitably be congestion. Essentially, this is the classic problem of ....

Claffy, K. C., Polyzos, G. C., and Braun, H.-W. (1992). Traffic characteristics of the t1 nsfnet backbone. Tech. rep. CS92-252, UCSD. Available via Merit gopher in Introducing the Internet directory.

....by this research (1988 through 1993) the NSFNET backbone gradually evolved from a T1 speed (1.544 Mbps) to a T3 speed (44.736 Mbps) network. Each backbone node is connected to several other nodes, typically 2 4. The nodes are responsible for packet switching, routing, and data collection [1]. The most dominant feature of the backbone traffic over the last few years is the overall increase in volume. While a rough estimate of future traffic might be obtained by simply fitting a smooth curve to the data, this method ignores a great deal of information. For example, the traffic may have ....

K. Claffy, G. C. Polyzos, and H.-W. Braun, "Traffic Characteristics of the T1 NSFNET Backbone," Proceedings IEEE INFOCOM'93, pp. 885-892, March 28 - April 1, 1993.

....include distributions of queue length or available buffers in nodes. Several studies of local area environments focus on short term utilization characteristics [2] 23] 3] Longer term utilization metrics would include traffic volume growth over several years on a backbone infrastructure [24]. An Internet service provider will tend to pay attention to utilization metrics as indicators of how close their network is to saturation so they can plan for upgrades. 2.4.2 Reachability As networks increase their range of possible destinations, so does the size of routing tables, and thus the ....

.... train model to study traffic locality behavior on a local area network [28] Other studies, though not focused on packet trains in particular, also find evidence for locality even in 10 networks of wider geographic scope, e.g. regional networks and national backbones [29] 5] 30] 31] 32] [24] [7] Others [33] 34] 35] have extended the packet train model to the transport and application layers, defining a train as a quadruple of source destination address pairs in conjunction with port numbers. Using a transport layer definition of a flow rather than a packet train definition, ....

[Article contains additional citation context not shown here]

K. Claffy, G. C. Polyzos, and H.-W. Braun, "Traffic characteristics of the T1 NSFNET backbone," in Proceedings of IEEE Infocom 93, pp. 885--892, 1993.

....and behavior in order secure the integrity of the network. In this section we highlight several examples of the benefits of traffic characterization to the higher layers that end users care about. 3. 1 long range Internet growth tracking Few studies on national backbone traffic characteristics [7, 9, 10, 11, 12, 13] exist, limiting our insight into the nature of wide area Internet traffic. We must rely on WAN traffic characterization studies that focus on a single or a few attachment points to transit networks to investigate shorterterm aspects of certain kinds of Internet traffic, e.g, TCP [3] TCP and UDP ....

K. Claffy, G. C. Polyzos, and H.-W. Braun, "Traffic characteristics of the T1 NSFNET backbone," in Proceedings of IEEE Infocom 93, pp. 885--892, March 1993.

....at optimal levels of granularity. This research is of particular importance now since new services not only dramatically increase network traffic, but can also completely change traffic profiles. Our initial traffic characterization results, focusing on the T1 NSFNET backbone, appeared in [CPB93]. We are now studying the accuracy of various traffic sampling mechanisms for data collection; this is very important for high speed networks where the amount of traffic makes it almost impossible to capture complete traces. Other directions pursued include characterization of international ....

Claffy, K., Polyzos, G. C., and Braun, H. W., "Traffic Characteristics of the T1 NSFNET Backbone, " Proc. IEEE INFOCOM '93, San Francisco, CA, March 28 - April 1, 1993.

First 50 documents

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