摘要:
Techniques and systems for server farm load balancing and resource allocation are disclosed. In one embodiment, a method of load balancing can include: arranging servers into service groups; receiving an access request with information related to a differentiation between the service groups; selecting one of the service groups based on a mapping comparison to the information; and selecting one of the servers within the selected service group based on a hardware utilization comparison. The servers can include GPRS (General Packet Radio Service) Gateway Support Node (GGSN) or Remote Authentication Dial In User Service (RADIUS) servers, for example. The information can include an Access Point Name (APN) or Calling Station ID, for example.
摘要:
Techniques and systems for server farm load balancing and resource allocation are disclosed. In one embodiment, a method of load balancing can include: arranging servers into service groups; receiving an access request with information related to a differentiation between the service groups; selecting one of the service groups based on a mapping comparison to the information; and selecting one of the servers within the selected service group based on a hardware utilization comparison. The servers can include GPRS (General Packet Radio Service) Gateway Support Node (GGSN) or Remote Authentication Dial In User Service (RADIUS) servers, for example. The information can include an Access Point Name (APN) or Calling Station ID, for example.
摘要:
Techniques for distributing network traffic from an access server to a service gateway include receiving, at a load balancer, sticky table data that indicates an association between a particular subscriber IP address and a particular subscriber-aware service gateway in a gateway cluster. An input data packet is received with an input source address and an input transport-layer destination. If it is determined that the input transport-layer destination indicates a type of payload that uses a service gateway, then the particular service gateway associated with the particular subscriber is determined based on the sticky table and IP address in the input source address. An output data packet is directed to the particular service gateway using a link-layer or networking-layer destination address. These techniques allow a load balancer to be located anywhere on the network and to bypass a subscriber-aware service gateway for some data traffic.
摘要:
Techniques for distributing control plane traffic, from an end node in a packet switched network to a cluster of service gateway nodes that host subscriber-aware application servers, include receiving a control plane message for supporting data plane traffic from a particular subscriber. A particular service gateway node is determined among the cluster of service gateway nodes based on policy-based routing (PBR) for the data plane traffic from the particular subscriber. A message based on the control plane message is sent to a control plane process on the particular service gateway node. Thereby, data plane traffic and control plane traffic from the same subscriber are directed to the same gateway node, or otherwise related gateway nodes, of the cluster of service gateway nodes. This approach allows currently-available, hardware-accelerated PBR to be used with clusters of subscriber-aware service gateways that must also monitor control plane traffic from the same subscriber.
摘要:
Techniques for distributing control plane traffic, from an end node in a packet switched network to a cluster of service gateway nodes that host subscriber-aware application servers, include receiving a control plane message for supporting data plane traffic from a particular subscriber. A particular service gateway node is determined among the cluster of service gateway nodes based on policy-based routing (PBR) for the data plane traffic from the particular subscriber. A message based on the control plane message is sent to a control plane process on the particular service gateway node. Thereby, data plane traffic and control plane traffic from the same subscriber are directed to the same gateway node, or otherwise related gateway nodes, of the cluster of service gateway nodes. This approach allows currently-available, hardware-accelerated PBR to be used with clusters of subscriber-aware service gateways that must also monitor control plane traffic from the same subscriber.
摘要:
In one embodiment, a method for providing an ACK packet while queuing data is provided. One or more packets in a series of packets may be received from a client at a gateway. The gateway determines that a packet in the series of packets has not been received. The one or more packets are then queued. The queued packets may have included an ACK for one or more previously sent packets. Thus, if the gateway had forwarded the one or more packets that are queued, then the ACK would have been received by the server. However, the one or more packets are queued and thus the ACK included in the packets is not sent with the queued packets. Even though the plurality of packets are queued, an acknowledgement packet is generated and then sent for the previously sent packets. The acknowledgement packet acknowledges to the server that the one or more previously sent packets were received by the client. The server may then continue transmitting packets to the client and knows it does not need to retransmit the one or more previously sent packets.
摘要:
In one embodiment, a method for providing an ACK packet while queuing data is provided. One or more packets in a series of packets may be received from a client at a gateway. The gateway determines that a packet in the series of packets has not been received. The one or more packets are then queued. The queued packets may have included an ACK for one or more previously sent packets. Thus, if the gateway had forwarded the one or more packets that are queued, then the ACK would have been received by the server. However, the one or more packets are queued and thus the ACK included in the packets is not sent with the queued packets. Even though the plurality of packets are queued, an acknowledgement packet is generated and then sent for the previously sent packets. The acknowledgement packet acknowledges to the server that the one or more previously sent packets were received by the client. The server may then continue transmitting packets to the client and knows it does not need to retransmit the one or more previously sent packets.
摘要:
A network is disclosed that includes a first tier of forwarding agents connected to a first tier of network devices. A second tier of forwarding agents is connected to a second tier of network devices. A service manager is configured to receive a packet from one of the forwarding agents; determine the tier of the forwarding agent; and send an instruction to the forwarding agent directing the forwarding agent to forward the packet to a network device connected to the tier of forwarding agents that includes the forwarding agent.
摘要:
A method for communicating in a loadbalancing environment is provided that in a particular embodiment includes receiving a request packet from a network access server (NAS) to initiate a communication session. The request packet is then communicated to a tunneling protocol network server (TPNS) and a response packet is received in response to the request packet. The response packet establishes a tunnel that facilitates the communication session and that includes an identification element associated with the TPNS such that a data transfer associated with the communication session is executed between the NAS and the TPNS.
摘要:
An improved client network address translation (“NAT”) system is provided. Methods are provided for initializing connection control blocks with the client NAT addresses when a storage pool of connection control blocks are allocated. This avoids the costly overhead as connections are created and destroyed. Unlike the prior art solutions, once the connection control blocks of the present invention are allocated, they remain in effect until the storage subpool is deleted. Secondly, in order to provide as large a pool of client NAT addresses as possible, methods are provided for using intervals to allocate and reclaim client NAT addresses as subpools are allocated and freed.