Embedded Analytics
Embedded analytics agents are compact software elements that can be directly integrated into virtually any system or product, performing real time monitoring and analysis of the system, or the data passing through it, and reporting in real time to a management application. Embedded monitoring has been used for many years however during the last decade there has been a dramatic increase in the level of sophistication of analytics technology and the quality and detail of the metrics reported.
It is key that embedded agents are compact, portable and use very little CPU and memory, as they should have no discernable impact on the system in which they are embedded. For example Telchemy's VQmon uses approximately 0.0001 MIPS to compute accurate and detailed analytics for a Voice over IP call when integrated into a VoIP gateway, IP phone or mobile application.
Key parameters are often strongly time varying, which means that sampled metrics (e.g. average, min, max) may be uninformative or misleading; sending frequent sampled values is one solution however this results in a large amount of reporting traffic. Advanced analytics agents use statistical models that are able to accurately represent time varying data and allow the agent to report less frequently but still provide a high resolution view of performance. VQmon uses a multistate Markov model to accurately model time varying packet loss, which provides deep insight into the behavior of the network and allows accurate user perceived quality (QoE) to be measured in real time.
Systems are dynamic in nature and it is important to understand the value of measured parameters within the context of the current system state. For example a system may incorporate dynamic buffer management or adaptive loss concealment, and the impact of an out-of-range parameter may depend strongly on the state of the adaptive elements of the system. For example a VoIP system typically incorporates an adaptive playout buffer, the impact of variation in packet arrival time depends entirely on the current playout buffer level, hence an embedded VoIP analytics agent needs to view packet delay variation in the context of the playout buffer.
Real time IP based video, audio and speech present many challenges to embedded analytics as they are strongly impacted by time varying network impairments and have numerous adaptive elements; in addition to adaptive playout buffers, multimedia communications systems incorporate dynamic codecs that can frequently change mode or rate, and may also contain loss correction or concealment. This requires embedded analytics software that is able to measure multiple parameters in real time, understand how measured parameters interact with adaptive system components and compute meaningful system or application performance and user perceived quality.
It is key that embedded agents are compact, portable and use very little CPU and memory, as they should have no discernable impact on the system in which they are embedded. For example Telchemy's VQmon uses approximately 0.0001 MIPS to compute accurate and detailed analytics for a Voice over IP call when integrated into a VoIP gateway, IP phone or mobile application.
Key parameters are often strongly time varying, which means that sampled metrics (e.g. average, min, max) may be uninformative or misleading; sending frequent sampled values is one solution however this results in a large amount of reporting traffic. Advanced analytics agents use statistical models that are able to accurately represent time varying data and allow the agent to report less frequently but still provide a high resolution view of performance. VQmon uses a multistate Markov model to accurately model time varying packet loss, which provides deep insight into the behavior of the network and allows accurate user perceived quality (QoE) to be measured in real time.
Systems are dynamic in nature and it is important to understand the value of measured parameters within the context of the current system state. For example a system may incorporate dynamic buffer management or adaptive loss concealment, and the impact of an out-of-range parameter may depend strongly on the state of the adaptive elements of the system. For example a VoIP system typically incorporates an adaptive playout buffer, the impact of variation in packet arrival time depends entirely on the current playout buffer level, hence an embedded VoIP analytics agent needs to view packet delay variation in the context of the playout buffer.
Real time IP based video, audio and speech present many challenges to embedded analytics as they are strongly impacted by time varying network impairments and have numerous adaptive elements; in addition to adaptive playout buffers, multimedia communications systems incorporate dynamic codecs that can frequently change mode or rate, and may also contain loss correction or concealment. This requires embedded analytics software that is able to measure multiple parameters in real time, understand how measured parameters interact with adaptive system components and compute meaningful system or application performance and user perceived quality.