Measured performance of DICOMTCP and DICOMISO implementations

The emerging Digital Imaging and Communications in Medicine (DICOM) 3.0 Standard supports two distinct network profiles for the communication of medical information. As specified in Part 8 of the DICOM Standard, DICOM application layer functionality may be supported by a TCPIIP based stack plus the DICOM Upper Layer for TCP/IP, or one of several ISO International Standard Profiles (e.g., U.S. GOSH'). Both protocol stacks must support ISO Association Control Service Element (ACSE) service interface. The physical and medium access control layers of the protocol are essentially unspecified. With the support of the Radiological Society of North America (RSNA), the Electronic Radiology Laboratory at the Mallinckrodt Institute of Radiology has implemented a subset of DICOM over the TCP/IP stack. Portions of both the Composite and Normalized DICOM Message Service Element protocols are included. In parallel the Radiologic Computing and Imaging Science Laboratory at Penn State has implemented a subset of DICOM using SunNet/OSI for Transport and Network Layers and ISODE version 8.0 to supply Session, Presentation and ACS. Care has been taken in both implementations to provide test points at comparable layers of the protocol stack to enable performance measurements. Utilizing each DICOM implementation one pair of Application Entities was implemented, consisting of a Service Class Provider and a Service Class User. These Application Entities implement the DICOM Storage Service Class for CR, CT and Secondary Capture Service-Object pair Classes, exercising the DICOM Composite Services. To characterize the performance of both DICOM implementations, software probes were inserted into the code at appropriate places. A probe consists of a minimal code segment that records a time stamp, probe location and quantity of data passing its location. By strategically locating probes it is possible to measure both transfer rates and queueing times. Probes were located at the Transport Service interface to measure relative memory to memory transfer times. The performance of the DICOM Upper Layer Protocol for TCP/IP was also compared to that of ISO Session, Presentation and ACS. In all cases the DICOM default Little-Endian Transfer Syntax was utilized. A test suite of 10 CT images (512 x 512 x 12 bits) and 10 digitized radiographs (2K x 2.5K x 12 bits) were used for throughput tests. Minimum throughput times, maximum delays and maximum delay variations will be presented as well as relative performance at each protocol layer and throughput as a function of negotiated maximum PDU size.