|
|
|
|
 |
Using FocalPoint in Telecom Load DistributionTraditional Telecom Servers contain standard CPU blades running applications such as intrusion detection, VPN routing and spam filtering. However, although they offer flexibility and ease of programming, standard CPU subsystems or server blades do not have the required processing power to perform deep inspection of packet headers at 10Gb line rates and above. To solve this problem, today’s high performance Telecom Servers employ specialized CPUs or NPUs to inspect packets at line rate. These specialized processing cards are much more costly than commodity CPU blades. The FocalPoint devices provide and alternative approach using load distribution to enable full line rate processing using standard CPU blades.
A load distribution configuration using two members of the FM4000 Series product line, the FM4224 and FM4410, is shown in the figure above. Here, the packets arrive at the FM4224 on the network card and are distributed across multiple passive backplane links to other cards in the system. Load distribution can be done such that the deep packet processing function is spread across multiple server blades reducing the processing power required on each blade. Other server blades can be used for further service processing, providing a very flexible deployment of CPU resources. For some applications, specialized services cards can also be used. Packet Flow The figure below shows a simplified configuration, which can be used to illustrate packet flow through the system. Here we assume that link aggregation group (LAG) A connects to network A and LAG B connects to network B. For example, in a Telecom Server application, data may be processed as it flows between these two networks. A typical system may include four of these switches and up to 16 application blades with some of the blades used for packet inspection and some blades used for services. At the system ingress, packets will be distributed by the FM4224 line switches to one of the application blades. The FM4410 can then load balance the traffc to one of the CPUs on the application blade. The CPU can then determine if the packet needs further processing by services running on other blades. If so, the packet can be routed back through the line switches for further processing. When packet processing is complete, the egress packets are distributed across the line switches by the FM4410 device on the application blade. This in effect, distributes the packet across one of the egress LAG members back to the network. The load distribution uses a modified Persons hash function based on information in the packet header that defines a flow. FocalPoint can be configured to use various L2/L3/L4 header fields. A given flow will always use the same path through the fabric even if the source and destination addresses are reversed. See the load balancing documents listed in the left column for more information.
With perfect distribution across four application blades, each blade can process the incoming packets at one quarter the line rate which greatly reduces the processing requirements and allows the use of commodity CPU blades. Higher performance NPU subsystems can always be used in some blade slots if needed. Load distribution also provides high availability to the fabric allowing continued operation during application blade or network card failure. If a link failure is detected, software can reconfigure the LAG with the remaining members.
|
|
|
|
|
© 2008 Fulcrum Microsystems, Inc. |
|
