With the recent release of the CG6565(C) and CG 6060(C) digital PSTN and IP media processing boards, NMS has introduced powerful new platforms for building advanced telephony applications in TDM and IP networks. These boards have replaced the popular CG 6000, CG 6000C, and CG 6100C.
One of the challenges in designing VoIP applications with any media processing board is predicting the IP session density, due to the very wide range of application behavior that is possible. This document will help developers better understand the VoIP performance characteristics of the CG 6565(C) and CG 6060(C).
Figures 1 and 2 show comparative performance against application benchmarks for the CG 6000 and CG 6565 PCI and CompactPCI boards. The greater processing power of the CG 6565 and CG 6565C boards is evident in the greater density of VoIP ports.
CG 6565(C) Performance Comparison
For the CG 6565(C) application benchmarks, VoIP streams were generated with a 20 msec packet size. In all cases a “port” or VoIP session is a full-duplex connection. A G.711-to-G.729A gateway performs real-time full-duplex transcoding between two VoIP endpoints. The G.729A IVR application plays a prompt and then relays transcoded voice to a speech recognition server. The G.711 Conferencing application connects three parties for each conference, so the reported session capacity represents the total number of conference callers. For G.711 Announcements, the application simply plays a file without processing the received audio stream.
Note that the comparisons assume that the boards support the same number of MIPS.
Figure 1: CG 6565 PCI Board Performance
Figure 2: CG 6565 PCI Board Performance
CG 6000(C) vs. CG 6060(C) Performance Comparison
For CG 6060(C) application benchmarks, VoIP streams were generated with a 20 msec packet size. In all cases, a "port" or VoIP session has its capacity recorded for identical benchmark tests.
Note that the comparisons assume that the boards support the same number of MIPS.
Figure 3: Selected CG 6060(C) Board Performance Benchmarks
Performance Enhancing Features of the New CG Boards
The significant performance gains for VoIP applications running on the CG 6565(C) and CG 6060(C) are mostly due to the fast PowerPC processor controller replacing the CG 6000’s StrongArm processor, but many other features are contributing factors (see Figure 4).
Figure 4: Comparison of the CG Series Boards (click to enlarge)
Compared to the CG 6000 and CG 6500C boards, the DSP processors have been updated to 30 percent faster parts with double the density. For example, this allows the CG 6565C to provide up to 12,800 MIPS per board and achieve up to six G.729 vocoder instances per DSP core.
Also of note on the CG 6565(C) boards, Ethernet ports may be configured for Gigabit speed. In addition, the PCI interface is now compatible with PCI-X and can accommodate bus configurations up to 133 MHz and 64-bit data paths.
The CG 6565(C) and CG 6060(C) employ a new approach to echo cancellation — a hardware-based echo canceller is available on all trunk lines, so that each input DS0 channel can have G.168-compliant echo cancellation with 64 msec tails. This is a significant improvement over earlier CG boards, not only because of improved algorithm quality and tail length, but because the DSP chips are no longer required to perform echo cancellation, allowing that DSP capacity to be used for other purposes.
Another DSP optimization on the new CG boards is the processing of call signaling, which has also been moved from DSP-based processing to hardware. Call signaling processing could consume up to two entire DSP cores, so removing this load also frees more DSP capacity for other functions.
A welcome feature for many developers is the incorporation of a fully non-blocking timeslot interchanger. Programming TDM circuit connections on older CG boards could sometimes lead to frustrating situations in which a call couldn’t be completed because there weren’t enough timeslot switching resources. The CG 6565(C) and CG 6060(C) solve this problem by allowing any TDM resource connection to be made without restriction.
Many applications using VoIP will greatly benefit from the new Natural Access™ feature called Native RTP. Without this feature, recording a message from a VoIP caller with a complex vocoder type involves a significant amount of DSP processing in order to decode the RTP stream and then compress the audio for storage. This "double-transcoding" process also degrades audio quality. The Native RTP feature allows an RTP stream to be directly recorded in its original compression format, without the cost and quality degradation associated with double-transcoding. This feature can also be used to reduce the cost of prompt playing by pre-recording prompts in the Native RTP format so that real-time transcoding of the prompt is not necessary. Refer to the "Using Native Play and Record" section of the "Fusion Developer’s Manual" for more details.
Figure 5: Features of the CG 6565 (click to enlarge)
Figure 6: Features of the CG 6060 (click to enlarge)