Public home page Local home page How to contact us Search

Catalog Description

Objective

Prerequisites

Topics

1. Design of modern processor architectures (3 weeks)

   Introduction to pipelining, instruction scheduling, branch processing, out of order execution, dealing with dependences. Current processor architectures are used to illustrate concepts.

2. Uniprocessor memory hierarchy design (2 weeks)

   Principals of memory hierarchy design including design of multi-level caches and virtual memory systems.

3. Multiprocessor design (4 weeks)
   • Snoopy cache based shared memory architectures
   • Sequential consistency, weak consistency, release consistency
   • Managing a multiprocessor memory hierarchy: tolerating and avoiding latency
   • Hardware and software support for distributed shared memory
4. High performance networks (4 weeks)

   Backplane, peripheral I/O channel, local area networks, wide area networks

   • Design of high bandwidth, low latency multiprocessor networks
   • HIPPI, Fibre Channel and Asynchronous Transfer Mode (ATM)
   • Local area network software, hardware able to support closely coupled workstation networks
5. I/O subsystems (4 weeks)

   Design and modeling of disks, disk caches, Redundant Arrays of Inexpensive Disks (RAID) devices. Parallel I/O subsystems, parallel file systems. Flash memory.

Course Text

1. Shlomo Weiss and James E. Smith. POWER and PowerPC, Morgan Kaufmann, 1994.
2. Craig Partridge, Gigabit Networking, Addison-Wesley, 1994.

Typical Grading and Workload