We were discussing Riverbed Operating System (RiOS) technical concepts. Today we look at QoS features of the product. We already discussed the network deployment and management streamlining options and found the RiOS devices to be appliances that expect no changes to the network. Now beyond the topology network considerations, let us review how QoS can be leveraged for bandwidth sensitive traffic. Traffic can be classified using the DS field (DSCP) This can be applied to both the optimized and pass-through traffic. The DS field values are passed by RiOS in completely transparent handoff and hence existing QoS can be honored. In addition, new QoS DS fields can be defined with the WAN accelerated traffic.
The data-deduplication and data reduction achieved with RiOS reclaims bandwidth and even undo existing QoS classifications which may no longer be necessary and can be replaced with simpler better re-classification. The RiOS devices that are deployed on the edge network can employ Hierarchical Fair Service class-of-service marking (H-FSC) and enforcement. This is used to improve both bandwidth and latency. H-FSC is a scheduling algorithm that can simultaneously support 1) hierarchical link sharing services, 2) guaranteed real-time service with provable tight delay bounds and 3) decoupled delay and bandwidth allocation. The latency-sensitivity of this approach is usually not found in many other techniques. If there are two different real-time critical applications and their bandwidth guarantees are met, the queues could still become filled with traffic introducing jitter. In such cases, the ability to schedule application traffic based on latency and setting priority to that application traffic comes very useful with this kind of a scheduling algorithm.
Application visibility into network traffic is improved with the Riverbed "AppFlow" classification engine which utilizes a variety of technique and usually in combination. These techniques include :
port - based classification for applications
pattern-matching or application signature using patterns or magic numbers in protocol headers using regular expressions, byte or string matching.
protocol dissection - which involves detailed interpretation of the application protocol for contextual sub-classification or protocol attribute extraction
future - flow registration - where a past occurrence of a flow is used to tag a future occurrence
behaviorial classification - where the behavioral classification relies on the detection of behavioral attributes of the network traffic using packet size, packet inter-arrival time, packet rate, data rate, and entropy calculations for a behavioral signature for an application.
decryption / decoding - where encoding, obfuscation or simple encryption may be used by applications
The data-deduplication and data reduction achieved with RiOS reclaims bandwidth and even undo existing QoS classifications which may no longer be necessary and can be replaced with simpler better re-classification. The RiOS devices that are deployed on the edge network can employ Hierarchical Fair Service class-of-service marking (H-FSC) and enforcement. This is used to improve both bandwidth and latency. H-FSC is a scheduling algorithm that can simultaneously support 1) hierarchical link sharing services, 2) guaranteed real-time service with provable tight delay bounds and 3) decoupled delay and bandwidth allocation. The latency-sensitivity of this approach is usually not found in many other techniques. If there are two different real-time critical applications and their bandwidth guarantees are met, the queues could still become filled with traffic introducing jitter. In such cases, the ability to schedule application traffic based on latency and setting priority to that application traffic comes very useful with this kind of a scheduling algorithm.
Application visibility into network traffic is improved with the Riverbed "AppFlow" classification engine which utilizes a variety of technique and usually in combination. These techniques include :
port - based classification for applications
pattern-matching or application signature using patterns or magic numbers in protocol headers using regular expressions, byte or string matching.
protocol dissection - which involves detailed interpretation of the application protocol for contextual sub-classification or protocol attribute extraction
future - flow registration - where a past occurrence of a flow is used to tag a future occurrence
behaviorial classification - where the behavioral classification relies on the detection of behavioral attributes of the network traffic using packet size, packet inter-arrival time, packet rate, data rate, and entropy calculations for a behavioral signature for an application.
decryption / decoding - where encoding, obfuscation or simple encryption may be used by applications
No comments:
Post a Comment