|
[October 18, 2001]
Synchronization Essentials Of VoIP
BY PAUL SKOOG AND DOUG ARNOLD
As we accelerate into the new world of voice over IP (VoIP), we often
assume we can leave some of the trappings of wireline telecom behind, such
as the need for synchronization. After all, an IP network is about as
asynchronous as it gets. While this may be true in some respects, precise
time and synchronization continue to permeate many areas of IP telephony
operations. Customer expectations of voice quality and service reliability
remain unchanged, and as a result, the need for accurate timing remains
unchanged though it manifests itself in different ways. The bottom line is
that trying to design a VoIP network without considering network
synchronization is probably the shortest path to realizing that you need
it.
Voice Latency
The greatest challenge in implementing a carrier-class VoIP system is
meeting the very high standard for voice quality set by the traditional
Public Switched Telephone Network (PSTN) system. The same is not true in
wireless phones, where inferior quality is accepted as the price of
mobility.
Voice calls using VoIP technology can be made to another VoIP phone or
to a traditional phone on the PSTN. Since VoIP phones currently represent
only a small fraction of phones in use, we will consider the latter case
in detail. The latency of the voice signal from the VoIP phone to the PSTN
phone consists of delays at the following network elements:
- The VoIP telephone;
- IP network routers or switches;
- The IP to PSTN gateway;
- The wires; or
- Other delays in the PSTN system.
At the phone, the signal must be sampled, encoded, and packaged as Real
Time Protocol (RTP) packets. Any routers encountered in the IP network
contain input and output buffers. At the gateway, the packets will
encounter more buffers (including the jitter buffer), plus delays
associated with decoding and reassembling the signal. The transmission
delay alone, due purely to travel time through wiring, would be about 20
ms for a call between Los Angeles and New York.
Lastly, travel through the PSTN system could involve breaking the
signal up into ATM or frame relay packets and reassembling them after
transport through an optical fiber. In this case, input and output buffers
are encountered again. If the final destination is a VoIP phone on another
network, then the phone call will have to be converted back into RTP
packets, complete with encoding, decoding, and buffer delays.
Latency Measurement
Essential to assuring low latency and acceptable quality of service (QoS)
and service level agreement (SLA) compliance, is the ability to make a
reasonable measurement of the latency in the network. This requires
adequate synchronization between measurement probes to obtain meaningful
results. Some measurement solutions use hop-by-hop roundtrip metrics to
build statistics on latency, jitter, etc. While this is informative, it
simulates more of a piecemeal measurement approach rather than a phone
conversation between two parties.
Measurement of end-to-end one way path latency better reflects a true
QoS metric. An even better measurement is a one-way latency test performed
simultaneously from both ends to reflect an in-process phone conversation.
Synchronization on each end of the call is required for this type of test.
VoIP offers the promise of converged of voice and data, cost savings
through reduction of transportation costs, and new products and services.
This promise hinges in part on the customer adoption rates of VoIP, which
in turn hinges on acceptable levels of quality of service. In the absence
of bulletproof methods to measure, monitor, or assure quality,
over-provisioning bandwidth becomes the de facto solution. Aside from
increasing costs, over-provisioning only increases the probability --
without guaranteeing the quality -- of the transmission of VoIP traffic.
Network Management, Fault Diagnosis, And Recovery
Most IT organizations are measured on their ability to maintain
full-flow network operations. This reliability requirement increases by
orders of magnitude when you add business-related voice traffic to the
same network. Any VoIP problem must be avoided, averted, or, in a
worst-case scenario, minimized in an effort to keep business-critical
voice systems running. To do this, one of the absolutely essential
underpinnings is the accuracy of the server and router log files.
Every log file entry is time stamped. These time stamps establish the
"when" of an event and as a group allow the ordering of events.
Log files and subsequent reports allow you to use the log file data to
identify root cause problems within your network. Since server logs are a
compilation of information from different hosts, it is essential that the
time stamps be correct. If not, you will have difficulty ordering events
and troubleshooting root-cause problems. The more difficulty you have in
identifying a problem, the longer the QoS level of your VoIP system may
degrade, or worse yet, be non-operational.
Figure 1. Time-accurate
server and router log files play a key
role in troubleshooting and identifying root cause problems.
Time synchronization across network servers, routers, and network
devices is not a difficult endeavor. Using the well-established Network
Time Protocol (NTP, RFC 1305) and a reliable time source, such as a
dedicated network time server that references the GPS system,
synchronization of servers and network devices can be easily maintained.
In fact, many operating systems and network devices already incorporate
support for NTP.
NTP uses Coordinated Universal Time (UTC), which is the same worldwide.
The GPS satellite system is the most readily-available source for UTC time
in the world. By synchronizing your network to UTC, you remove one more
source of interoperability problems between your network and others. This
is important since VoIP traffic may transit many networks, requiring the
correlation of log files from various networks to solve a problem. Time
servers on the market today provide accurate and secure time and are able
to synchronize thousands of clients on the network.
Call Detail Records Need To Be In Sync
No discussion regarding time on a VoIP network would be complete without
mention of the obvious role time plays in billing. Call Detail Records
(CDR) provide information about call origination, destination, and
duration. Duration certainly includes the time stamp when the call was
initiated, and either the call duration or time the call was terminated.
Billing integrity will rely on the underlying time accuracy of the VoIP
CDR records. Without proper synchronization, the CDR accuracy will falter
and the billing system will inevitably come into question. This is
particularly critical when the CDR information is shared between carriers
and billing discrepancies require time-consuming mediation.
Synchronizing The Gateway Interface
Eventually, packets arrive at the gateway between the VoIP system and the
PSTN. The PSTN uses a very defined timing hierarchy for synchronization of
traffic on the network. PSTN voice packets must arrive in order, and with
low latency and jitter. This gateway also represents a change in the
general synchronization requirements: VoIP systems synchronize by way of
time stamps to aid in latency reduction and network log file integrity;
the PSTN uses synchronization to improve efficiency and data throughput.
Figure 2. Precise time is
required for VoIP;
precise frequency is required for PSTN.
Providing this synchronization requires a versatile time reference that
can supply the Stratum 1 level frequency reference for the PSTN and the
accurate time stamps for the VoIP side. Many of these VoIP/PSTN gateways,
also known as a softswitches, already employ NTP for accurate time
stamping. Stratum 1 level timing is already permeating the edge the PSTN
network by synchronizing customer premise ATM routers and switches. By
adding VoIP to the edge, it further increases the need to expand the
synchronization capabilities of the Stratum 1 timing clock. Again, a
quality GPS-referenced clock can support both the NTP and the Stratum 1
time and frequency requirements.
Sync Now Or Sink Later
Synchronization is usually far from top priority when establishing a VoIP
network. However, as soon as quality problems occur, the true value of the
synchronization system becomes very clear. QoS monitoring systems and
network diagnostic programs will ultimately drive the requirement for
synchronization across network routers, servers, and related devices.
These systems rely on log file accuracy and integrity for their metrics.
Without time stamp accuracy, an unacceptable amount of time will be spent
trying to resolve problems that could have been avoided, or cleared up
more efficiently. Good practices consider synchronization important enough
to incorporate up-front in system design, rather than later when trouble
occurs.
Paul Skoog is the product manager for IP Networking Products, and
Doug Arnold is the chief scientist for TrueTime,
Inc. TrueTime designs and manufactures precision time and
frequency products for network and telecommunications synchronization.
|
