In my November 2002
column, I highlighted key new building blocks for Internet telephony
products -- including moving telephony media processing tasks traditionally
done using DSP algorithms to the host CPU processor. I received some e-mails
asking for more detail on this trend. Iï¿½ve also noticed some hype and
controversy around ï¿½host media processingï¿½ (HMP) or ï¿½soft media processingï¿½
in trade magazines, so Iï¿½ve decided to devote an entire column to HMP.
Opinions on the impact of HMP range from promising a quick and complete
transition from the ï¿½DSP nightmareï¿½ or ï¿½software nirvanaï¿½ to comparing it
with some well-known fiascos. As I see it, the transition from DSP-based
media processing to HMP will be similar to the transition from transistors
to integrated circuits (ICs).
While HMP technology still covers only a subset of the media serverï¿½s
needs -- on the lower end of the density and features range -- it can
already be cost-effectively used in some areas where the DSP-based
technology cannot. This is especially true in the voice-over-packet
environment, where standard Ethernet cards can be used for interfacing with
the network. This is why HMP can be called a disruptive technology. In the
long run, HMP will replace -- but not totally eclipse -- DSPs in a wider
range of media processing solutions.
Letï¿½s explore why and how the HMP phenomenon is happening now, and what
we can expect in the future.
WHY HMP IS DISRUPTIVE
HMP as a technology has existed for at least six years in ï¿½soft modems.ï¿½
However, this was just a sustaining improvement that made modems more cost
effective. It did not really create new usage scenarios or business models,
which happens when HMP is combined with voice over IP because it is:
ï¿½ Form factor independent. It can be used without modification with PCI,
Universal PCI, CompactPCI, Advanced TCA, and other form factors.
ï¿½ Packet network interface independent. It can be used without modification
with 10 Mb, 100 Mb, 1 Gb, or wireless Ethernet.
ï¿½ Highly flexible. HMP scales down to be configurable for just one port,
making it possible to use it in applications where using DSP boards is
ï¿½ A software-only product. HMP does not involve any installation,
configuration, and maintenance procedures for the hardware, which can change
business models for some ecosystem players.
Plus, performance of HMP is gaining even faster than Mooreï¿½s law, since
it is being optimized as it evolves.
COST EFFECTIVENESS IS KEY
With Mooreï¿½s Law expected to work for at least another decade (according to
a recent C-NET article), HMP technology has a significant growth potential.
The price/performance ratio of processors also continues to rapidly improve.
While the cost per MHz of a server (including all required hardware) was
around $10 to $12 in 1996, it dropped to approximately $1 by 2002. We can
expect it to drop below $0.5 by 2005 if the current trend continues. This
means the cost of a standard high-volume (SHV) processing platform for
building media servers will, at some point, become much more attractive than
specialized DSP-based hardware.
HMP is mainly attractive for solutions that must scale down, not up. For
example, it can be used to build IP media servers for the small and medium
enterprise with two, four, or eight ports per system. This is a market
segment where no existing IP media processing boards can be cost effectively
used. Over time, HMP will scale up (some vendors already claim 200 ports per
system) and expand into higher solution densities.
Another factor making HMP-based VoIP solutions easier to scale is that
HMP allows these solutions to be built in a distributed fashion using open,
standards-based interfaces such as SIP, MGCP, VoiceXML, and SALT.
Building HMP is not without challenges -- primarily in the area of
providing real-time performance while scaling up to hundreds of ports per
server on a general-purpose operating system like Windows or Linux. Meeting
these challenges requires special techniques such as running the software at
real time priorities in the kernel mode using high-resolution timers and a
high degree of code optimization.
So HMP will not completely replace DSP-based hardware, which will still
be used for solutions requiring high usage of processing resources (such as
transcoding between media encoding formats) and for large-scale solutions.
It will be similar to the way vacuum tubes were used (and are still used
today) even after transistors were introduced.
MATURING INTO THE FUTURE
Host media processing technology has matured to the point where it is ready
to be deployed in the mainstream media server solutions. Today, its main
value is on the lower end. Over the next few years it will continue
expanding into higher densities. Ultimately, while HMP will not completely
replace DSP, it will complement it for a wide range of media processing
Jim Machi is director, Product Management for the Network Processing
Division of the Intel Communications Group. Intel, the worldï¿½s largest
chipmaker, is also a leading manufacturer of computer, networking, and
communications products. For more information, visit
To The July 2003 Table Of Contents ]