Emilie Thorbjorg Saulnier

High level traffic analysis of a LAN segment

A data collection and analysis study of packet traffic over a local area network is reported. The objectives of this study are to develop analysis tools to isolate the effect of each protocol, and eventually each application. The measurements of the LAN are made using a raw Ethernet packet capture program (tcpdump) running on a Sun workstation. The analysis was accomplished using a PERL script and the ACE/gr data analysis program. The contribution of each protocol and host is isolated, and the higher-level application mix is examined. The characteristics of the protocol traffic are examined, and it is shown that the lower-rate traffic protocol characteristics, while not Poisson, are similar to a proposed two-state Markov chain model, while the higher-rate protocol/application traffic exhibits correlation and periodicity that are more difficult to model analytically.<<ETX>>

A partitioned implementation of the Xpress Transfer Protocol. I

This paper focuses on the practical aspects of implementing the Xpress Transfer Protocol (XTP). A strategy is proposed for partitioning XTP across host and off-board processors in a distributed system in which an existing XTP implementation is changed in a step-wise fashion to meet the system requirements. The architecture is described, and the XTP development strategy is presented. Finally, XTP performance and integration issues are discussed.<<ETX>>

Data transfer bottlenecks over SPARC-based computer networks

In this paper we report on a data collection and analysis study of data transfers over a SPARC-based computer workstation network. The objectives of this study are to identify the performance characteristics and bottlenecks of these types of transfers. Measurements of performance over both Ethernet and CDDI LANs are made using two application level benchmarking programs: SPIMS and TTCP. It is shown that for earlier workstation models the performance bottleneck is not the network, while more recent models can more effectively utilize the network. This reflects the fact that perceived problems with network performance are often due to the limitations of the computer resources such as memory, rather than the network protocol itself. These types of measurements, in combination with other approaches, can help to reliably design, and predict performance of, distributed network applications.

High Level Analysis of Medical Imaging Traffic

Medical network traffic is characterized by the mix of modalities and applications on the network. Since different modalities or applications have different characteristics on the network, understanding the mix of traffic loads is important in analyzing network activity. Although many commercial products can analyze network traffic at the packet level, they do not capture higher level modality load characteristics or isolate the impact of large image transfers. More insight can be obtained through a high-level analysis of this packet data in which network traffic is examined at an application level. In this paper we report on data collection and analysis of medical network traffic over a local area network. The measurements of the LAN are made using a raw ethernet packet capture program (tcpdump) running on a Sun workstation, and the analysis was accomplished using a PERL script and the ACE/gr data analysis program. The contribution of each modality is isolated and the network protocol performance is examined in detail.

Modeling heterogeneous sources on multiple time scales

We develop parameters and results for a set of traffic models which emulate the traffic patterns at the packet and buffer levels. We show that traffic models at these levels are dependent on the network capacity as well as source-induced bottlenecks such as buffer reads and writes. We propose a pragmatic approach which uses three variations of a simple two-state Markov superposition model to describe the buffer and packet level view of the traffic under different bottleneck conditions. We apply these models to actual network traffic sources as measured on a medical network and evaluate the impact of changing bottlenecks in network transfers. Our data and results emphasize the sensitivity of any traffic model to the underlying machine capabilities and the timescale of interest. These models can be used to characterize a wide range of traffic types, and allow relatively simple analytic and simulation implementation.

Experience with a proposed teleradiology system for digital mammography

Teleradiology offers significant improvement in efficiency and effectiveness over current practices in traditional film/screen-based diagnosis. In the context of digital mammography, the increasing number of women who need to be screened for breast cancer, including those in remote rural regions, make the advantages of teleradiology especially attractive for digital mammography. At the same time, the size and resolution of digital mammograms are among the most challenging to support in a cost effective teleradiology system. This paper describes a teleradiology architecture developed for use with digital mammography by GE Corporate Research and Development in collaboration with Massachusetts General Hospital under National Cancer Institute (NCI/NIH) grant number R01 CA60246-01. Experience with a testbed prototype is described. The telemammography architecture is intended to consist of a main mammography diagnostic site serving several remote screening sites. As patient exams become available, they are forwarded by an image server to the diagnostic site over a WAN communications link. A radiologist at the diagnostic site views a patient exam as it arrives, interprets it, and then relays a report back to the technician at the remote site. A secondary future scenario consists of mobile units which forward images to a remote site, which then forwards them to the main diagnostic site. The testbed architecture is based on the Digital Imaging and Communications in Medicine (DICOM) standard, created by the American College of Radiology (ACR) and National Electrical Manufacturers Association (NEMA). A specification of vendor-independent data formats and data transfer services for digital medical images, DICOM specifies a protocol suite starting at the application layer downward, including the TCP/IP layers. The current DICOM definition does not provide an information element that is specifically tailored to mammography, so we have used the DICOM secondary capture data format for the mammography images. In conclusion, experience with the testbed is described, as is performance analysis related to selection of network components needed to extend this architecture to clinical evaluation. Recommendations are made as to the critical areas for future work.

Experience with an XTP implementation for embedded systems

An implementation of the XTP protocol is discussed relative to the embedded systems environment. An overview of the embedded system architecture and unique design constraints is followed by a description of the protocol development environment. The original development approach is presented with a critique based on experience thus far. Embedded system implementation and protocol development issues experienced in this type of environment and proposed solutions are presented to provide a framework for similar development efforts.<<ETX>>

Simplified ethernet model for large-image transfer

In creating a higher level ethernet model the important effects of the ethernet bus on a PACS system must be taken into consideration. These effects include the delay in time between the submission of an image for transmission and the time the last packet arrives at the receiving node, and the retransmission of packets due to collisions. A good image-level model should approximate the statistical characteristics of the image delay over a range of network loading. It should also predict the number of packets which will require retransmission and the statistical characteristics of this effect over a range of loading. This paper describes the approach used to create such a model and compares the results obtained when using the BONeS model at the packet and image levels. It is shown that the approximate image-level ethernet model is much faster than the packet-level simulation while also providing accurate results. The impact of packet loading time on ethernet fairness is also discussed.