A WCCP router automatically distributes requests over all NetCache appliances that are in the same WCCP service group as the router.
Configuration of service group information on a NetCache appliance enables the appliance to send HERE_I_AM messages identifying the services it supports, for example, client-side TCP port 80 traffic (HTTP). Corresponding configuration of service group information on the router enables it to redirect and distribute traffic for services it has been configured to support. A WCCP router listens for HERE_I_AM messages to determine the existence of proxy-cache servers and the services each supports.
For example, assume that a WCCP router is configured to listen for HTTP and NNTP requests. When it detects a NetCache appliance in WCCP service group 0 sending HERE_I_AM messages, it determines that it can send TCP port 80 (HTTP) requests to that appliance. The router uses its hashing function to determine which appliance in service group 0 to send the request to.
If a particular NetCache appliance is not sending HERE_I_AM messages because it is unavailable, the appliance is no longer in the router's list of available appliances over which the router can distribute traffic at that time.
Part of WCCP service group definition in NetCache is identifying the hashing function the router will use to distribute requests across proxy-cache servers in the same service group as the router. Typically, you will want to achieve an even distribution of requests across your NetCache appliances. You can, however, add weighting factors to the basic request distribution criteria to achieve an uneven distribution of requests over your NetCache appliances, as described in Adjusting request distribution based on weighting.
The following table shows the basic criteria available for request distribution and how they apply to the modes in which you can run a NetCache appliance.
The details about the two distribution methods for IP address hashing are described in the following paragraphs.
When users are sending requests to arbitrary Internet servers (that is, requests are sent to many destinations), distribution on destination IP address provides the most equal distribution over multiple NetCache appliances that are running as forward proxies.
Requests for objects from the same origin server are sent to the same proxy-cache server, thereby minimizing duplication of objects among proxy-cache servers. When a proxy-cache server that was unavailable becomes available again, it takes over responsibility for handling requests for the same origin servers that it handled previously.
If you are deploying an accelerator, configuring the router to partition requests based on source IP address provides the best distribution over your NetCache appliances. Each appliance receives a portion of the load of client requests. The problem with partitioning requests based on destination IP address is this: if the NetCache appliance is accelerating only one origin server, all requests will be sent to one NetCache appliance because the destination address would be the same for all requests. (The hashing function sends requests for the same destination IP address to the same NetCache appliance.) Even if your NetCache appliance is accelerating a few origin servers, the variation in destination IP addresses would be limited.
For an accelerator, adding the additional criteria of hashing on source port as well as source IP address results in requests from the same client being distributed over multiple NetCache appliances. The reason is that the source port numbers are dynamically assigned by the operating system on the client and are in effect only for the life of the connection.
When partitioning is based on source IP address, objects might be duplicated among your accelerators because multiple NetCache appliances would be fetching the same objects. However, the hit rate with an accelerator is higher than with a forward proxy. Therefore, the duplication of objects is not a concern, as it is with a forward proxy. The reason is that the Web server or streaming server that NetCache accelerates has a limited amount of data, as compared to the World Wide Web, which has nearly an infinite amount of data. With an accelerator, it is likely that many users will send requests for the same data from the Web server or streaming server.
See Strategies for client access to an accelerator for an example of an accelerator deployment with a switch or WCCP router.
NetCache provides default weighting criteria to enable the WCCP router to factor in weighting with the basic request distribution criteria you provide. More requests would, therefore, be sent to a C6100 (a large model NetCache appliance) than to a C1100 (a small model). Typically, you would not change NetCache defaults for weighting. However, it might be beneficial to change NetCache defaults if your traffic patterns indicate that the default weighting does not achieve the most efficient request distribution for your organization.
Note
The weighting feature cannot be used with appliances running NetCache versions prior to 5.1 or with third-party proxy-cache servers.
If a NetCache appliance is overloaded, NetCache can refuse to accept packets for new connections, automatically returning them to the router. The router then directs those packets to the origin server.
NetCache maintains all connections initiated prior to overloading and refuses packets only for new connections. When the NetCache appliance begins to accept new connections again, a client connection timeout is initiated so that any new client connections (that might hang because they have been interrupted) are not left open indefinitely.
Typically, in a transparent deployment with a WCCP router, the NetCache appliance chooses one network interface over which to receive requests from clients. NetCache always delivers data to clients over the same interface over which it received the requests. However, delivery of data over a single network interface might not be optimal for some types of deployments.
For example, a streaming media cache can deliver streaming data to clients very quickly over multiple network interfaces. However, if the streaming media cache can deliver streaming data over only a single network interface, it is likely to become limited in its network throughput by the bandwidth of this single network interface.
Through NetCache configuration, you can identify the NetCache network interfaces over which you want the NetCache appliance to receive requests from clients, thereby enabling the appliance to deliver data over multiple network interfaces. In this case, the NetCache appliance appears to the WCCP router to be multiple proxy-cache servers.
Copyright © 1994-2002, Network Appliance, Inc. All rights
reserved.
Part No. 210-00372
File: access8.htm
Last updated: 11/13/02 18:10:53