Scalable Video Servers : PCI Bus Mointoring of Video Streams.

Problem

Building Scalable Video Servers brings Computer architects and Operating Systems researchers together.  Current areas of research include media storage, streaming of video streams, synchronization of video and audio streams and scheduling of streams. Our research focuses on the scheduling of media streams. To this end, we have implemented a host-based stream scheduler and a stream scheduler resident on the network interface. The problem is to be able to transfer streams from different producers to a Scheduler card resident on the I/O bus in an efficient manner. PCI peer-to-peer messaging, DMA transfers and PCI bus reads and writes are some ways of intracomputer stream transfer. The performance of these may however be limited by DMA transfer of a partial cache line ( dirty line in cache on target CPU), DMA stalls because of inefficent implementation of the DMA transfer engines and other issues. We would like to study these performance bottlenecks in video stream transfer.  We are currently using a Compaq Pamette (PCI card with 4  Xilinx 4K parts) for some of our research. A program may be downloaded to the Pamette ( C/VHDL compiled to Xilinx bitstreams ) for PCI
bus monitoring of streams
 

Here is what you need to do.

• Compile and download program for PCI bus monitoring of streams ( program already  available).
• Monitor PCI bus transfers between two i960 RD PCI cards.  One cards serves as a stream producer and the other serves as a consumer or scheduler.
• Design a scheme to be able to log  PCI transactions
• Determine possible PCI transfer bottlenecks
• Two page report detailing the above

• Knowlege of C. VHDL useful. FPGA and hardware background useful.
• PCI Bus SIG (Industry consortium)
• I2O Special Interest Group
• PCI Pamette Page

• Write a  2  page report addressing the following issues:
• Experimental setup
• Identify PCI bus bottlenecks
• PCI transaction logging facility