Browse Prior Art Database

Demand Data Cache Prefetching to Improve the Performance of Multimedia Applications

IP.com Disclosure Number: IPCOM000106806D
Original Publication Date: 1993-Dec-01
Included in the Prior Art Database: 2005-Mar-21
Document File: 6 page(s) / 195K

Publishing Venue

IBM

Related People

Chan, JC: AUTHOR [+5]

Abstract

Multimedia is by far the hottest technology around. Several vendors today are offering multimedia solutions to education, training, merchandising, and business presentation. Many of them are software-based, low-cost products running on PCs powered by Intel's 386 and 486 microprocessors. It has been observed that developing low-cost software multimedia solutions will continue to dominate a market share so long as chip manufacturers continue to deliver low-cost, high performance microprocessors.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 32% of the total text.

Demand Data Cache Prefetching to Improve the Performance of Multimedia Applications

      Multimedia is by far the hottest technology around.  Several
vendors today are offering multimedia solutions to education,
training, merchandising, and business presentation.  Many of them are
software-based, low-cost products running on PCs powered by Intel's
386 and 486 microprocessors.  It has been observed that developing
low-cost software multimedia solutions will continue to dominate a
market share so long as chip manufacturers continue to deliver
low-cost, high performance microprocessors.

      The performance of multimedia applications generally depends on
three factors: the audio/video file system access time, the network
bandwidth (in case of running applications over a network), and the
CPU processing time.  For software-only solutions, the CPU is
generally involved in decompressing the compressed data stream for
audio/video playback.  As an example, for playing back a 10-minute
full-motion video at the rate of 30 fps (frames per second) on a
resolution of 352x240 with 24 bits per pixel, a total of 4.6 GB
decompressed data needs to be re-generated, i.e., the data will be
re-generated at the bit rate of 60.8 Mbps (mega bits per second).
This is equivalent to the processing of a compressed video file of
size 112.5 MB which is encoded by employing the MPEG video
compression algorithm at the bit rate of about 1.5 Mbps.  Since most
of the multimedia applications are real-time (or at least time
guaranteed), any excessive delay incurred in decompressing the data
will cause severe damage to the sound and video quality.  Therefore
it is important to design a processor that can minimize the delay in
processing compressed audio/video data.

      Most of today's high performance RISC processors are designed
with an on-chip data cache to buffer up memory data that might be
re-used by the processor.  Since most of the compressed audio/video
files are very long and sequential, the decompression process is
basically required to walk through the entire file sequentially.  It
is conceivable that a high data cache miss ratio will be encountered.
For example, if a data cache is designed with a 4-byte interface,
then a 6.25% miss ratio will be experienced if each line contains 64
bytes of data.  This kind of miss ratio is not tolerable on any of
today's high speed machines.

      To reduce the data cache miss ratio while running multimedia
applications, we propose a new prefetching scheme which prefetches
the data based on the occurrence of multimedia data.  The basic idea
is this: while reloading the current line from the memory, the next
line will always be prefetched providing the current line being
reloaded contains multimedia data.  The reason for doing so is clear.
As mentioned earlier, the decompression algorithm is processing the
data SEQUENTIALLY through a com file.  Since the file size is much
larger than that of the dat...