12

Load Balancing a Cache Array for Remote Server Acceleration

• Introducing remote server acceleration
• Configuring remote server acceleration
• Creating pools
• Creating a cache rule
• Creating a virtual server
• Configuring for intelligent cache population
• Additional configuration options

Introducing remote server acceleration

This chapter explains how to set up a remote server acceleration configuration, in which a BIG-IP uses content-aware traffic direction to enhance the efficiency of an array of cache servers that cache content for a remote web server.

Note: All BIG-IP products except the BIG-IP LoadBalancer Controller support this configuration.

Figure 12.1 illustrates the remote server acceleration configuration, and provides an example configuration for this entire chapter. Remember that this is just a sample: when creating your own configuration, you must use IP addresses, host names, and so on, that are applicable to your own network.

Figure 12.1 Remote server acceleration

This configuration is similar to the configuration discussed in Chapter 11, Load Balancing a Cache Array for Local Server Acceleration. The difference is that, in this configuration, the cache servers reside on an intranet network, while the origin web server resides on the Internet; in the local server acceleration configuration, the origin web server and the cache servers all reside on the intranet. The remote server acceleration configuration is appropriate for any enterprise in which the cache server network and web server network are separated, and you want maximum speed and efficiency from both the cache servers and web server.

The configuration detailed in this chapter uses the following BIG-IP features:

• Cacheable content determination
  With cacheable content determination, you can determine the type of content you cache on the basis of any combination of elements in the header of an HTTP request.
• Content affinity
  Content affinity ensures that a given subset of content remains associated with a given cache to the maximum extent possible, even when cache servers become unavailable, or are added or removed. This feature also maximizes efficient use of cache memory.
• Hot content load balancing
  Hot content load balancing identifies hot, or frequently requested, content on the basis of number of requests in a given time period for a given hot content subset. A hot content subset is different from, and typically smaller than, the content subsets used for content striping. Requests for hot content are redirected to a cache server in the hot pool, a designated group of cache servers. This feature maximizes the use of cache server processing power without significantly affecting the memory efficiency gained by content affinity.
• Intelligent cache population
  Intelligent cache population allows caches to retrieve content from other caches in addition to the origin web server. This feature is useful only when working with non-transparent cache servers, which can receive requests that are destined for the cache servers themselves, as opposed to transparent cache servers, which can intercept requests destined for a web server. Intelligent cache population minimizes the load on the origin web server and speeds cache population.

Maximizing memory or processing power

From the time you implement a cache rule until such time as a hot content subset becomes hot, the content is divided across your cache servers, so that no two cache servers contain the same content. In this way, efficient use of the cache servers' memory is maximized.

After a hot content subset becomes hot, requests for any content contained in that subset are load balanced, so that, ultimately, each cache server contains a copy of the hot content. The BIG-IP distributes requests for the hot content among the cache servers. In this way, efficient use of the cache servers' processing power is maximized.

Thus, for a particular content item, the BIG-IP maximizes either cache server memory (when the content is cool) or cache server processing power (when the content is hot), but not both at the same time. The fact that content is requested with greatly varying frequency enables the cache statement rule to evaluate and select the appropriate attribute to maximize for a given content subset.

Configuring remote server acceleration

To configure remote server acceleration, complete the following tasks in order:

• Create pools
• Create a cache rule
• Create a virtual server
• Configure for intelligent cache population
  
  

  Each of the following sections explains one of these tasks, and shows how you would perform the tasks in order to implement the configuration shown in Figure 12.1. Note that in this example, as in all examples in this guide, we use only non-routable IP addresses. In a real topology, the appropriate IP addresses would have to be routable on the Internet.

Creating pools

To use the remote server acceleration configuration, you create three sets of load balancing pools. You create pools for your origin server (the web server on which all your content resides), for your cache servers, and for your hot, or frequently requested, content servers, which may or may not be cache servers. A pool is a group of devices to which you want the BIG-IP to direct traffic. For more information about pools, refer to Pools in the BIG-IP Reference Guide, Chapter 4, Configuring the High-Level Network.

You will create these pools:

• Cache server pool
  The BIG-IP directs all cacheable requests bound for your web server to this pool, unless a request is for hot content.
• Origin server pool
  This pool includes your origin web server. Requests are directed to this pool when:
  
  
  • The request is for non-cacheable content; that is, content that is not identified in the cacheable content expression part of a cache statement. For more information, see Working with a cacheable content expression, on page 12-5.
  • The request is from a cache server that does not yet contain the requested content, and no other cache server yet contains the requested content.
  • No cache server in the cache pool is available.
• Hot cache servers pool
  If a request is for frequently requested content, the BIG-IP directs the request to this pool.
  
  

  Note: While the configuration shown in Figure 12.1 implements a hot cache servers pool, this pool is not required if you want to use the content determination and content affinity features. However, you must implement this pool if you want to use the hot content load balancing or intelligent cache population features.

To create a pool using the Configuration utility

1. In the navigation pane, click Pools.
   The Pools screen opens.
2. Click the Add button.
   The Add Pool screen opens.
3. For each pool, enter the pool name and member addresses in the Add Pool screen. (For additional information about configuring a pool, click the Help button.)
   
   

   Configuration Notes

   First, create a pool for the cache servers. For example:

   Create a pool named cache_servers with members 10.10.20.4, 10.10.20.5, and 10.10.20.6. For each cache server you add to the pool, specify port 80, which means this cache server accepts traffic for the HTTP service only.

   Create a pool named origin_server with member 192.168.10.1 and specify port 80, which means the server accepts traffic for the HTTP service only

   Create a pool named hot_cache_servers with members 10.10.20.4, 10.10.20.5, and 10.10.20.6 and specify port 80, which means this cache server accepts traffic for the HTTP service only.

To create a cache server pool from the command line

To define a pool from the command line, use the following syntax:

b pool <pool_name> { lb_method <lb_method> member <member_definition> ... member <member_definition> }

To create the cache server pool, type:

b pool cache_servers { \

member 10.10.20.4:80 \

member 10.10.20.5:80 \

member 10.10.20.6:80 }

To create the origin server pool, type:

b pool origin_server { member 192.168.10.1:80 }

To create the hot pool, type:

b pool hot_cache_servers { \

member 10.10.20.4:80 \

member 10.10.20.5:80 \

member 10.10.20.6:80 }

Note: If you have the hot content pool and the cache servers pool reference the same nodes, it enables use of the intelligent cache population feature.

Creating a cache rule

A cache rule is a specific type of rule. A rule establishes criteria by which a BIG-IP directs traffic. A cache rule determines where and how the BIG-IP directs content requests in order to maximize the efficiency of your cache server array and of your origin web server.

A cache rule includes a cache statement, which is composed of a cacheable content expression and two attributes. An attribute is a variable that the cache statement uses to direct requests. It can also include several optional attributes.

A cache statement may be either the only statement in a rule, or it may be nested in a rule within an if statement.

Working with a cacheable content expression

The cacheable content expression determines whether the BIG-IP directs a given request to the cache server or to the origin server, based on evaluating variables in the HTTP header of the request.

Any content that does not meet the criteria in the cacheable content expression is deemed non-cacheable.

For example, in the configuration illustrated in this chapter, the cacheable content expression includes content having the file extension .html or .gif. The BIG-IP considers any request for content having a file extension other than .html or .gif to be non-cacheable, and sends such requests directly to the origin server.

For your configuration, you may want to cache any content that is not dynamically generated.

Using required attributes

The cache rule must include the following attributes:

• origin_pool
  Specifies a pool of servers that contain original copies of all content. Requests are load balanced to this pool when any of the following are true:
  
  
  • The requested content does not meet the criteria in the cacheable content condition.
  • No cache server is available.
  • The BIG-IP is redirecting a request from a cache server that did not have the requested content.
• cache_pool
  Specifies a pool of cache servers to which requests are directed in a manner that optimizes cache performance.

Reviewing optional attributes

The attributes in this section apply only if you are using the hot content load balancing feature.

Note: In order to use the intelligent cache population feature, the cache_pool and the hot_pool must either be the same pool, or different pools referencing the same nodes.

• hot_pool
  Specifies a pool of cache servers to which requests are load balanced when the requested content is hot.

  The hot_pool attribute is required if any of the following attributes is specified:

• hot_threshold
  Specifies the minimum number of requests for content in a given hot content set that causes the content set to change from cool to hot at the end of the period.
  If you specify a value for hot_pool, but do not specify a value for this variable, the cache statement uses a default hot threshold of 100 requests.
• cool_threshold
  Specifies the maximum number of requests for content in a given hot content set that causes the content set to change from hot to cool at the end of the hit period.
  If you specify a variable for hot_pool, but do not specify a value for this variable, the cache statement uses a default cool threshold of 10 requests.
• hit_period
  Specifies the period in seconds over which to count requests for particular content before determining whether to change the content demand status (hot or cool) of the content.
  If you specify a value for hot_pool, but do not specify a value for this variable, the cache statement uses a default hit period of 60 seconds.
• content_hash_size
  Specifies the number of units, or hot content subsets, into which the content is divided when determining whether content demand status is hot or cool. The requests for all content in a given subset are summed, and a content demand status (hot or cool) is assigned to each subset. The content_hash_size should be within the same order of magnitude as the actual number of requests possible. For example, if the entire site is composed of 500,000 pieces of content, a content_hash_size of 100,000 would be typical.
  If you specify a value for hot_pool, but do not specify a value for this variable, the cache statement uses a default hash size of 1028 subsets.

Understanding content demand status

Content demand status is a measure of the frequency with which a given hot content subset is requested. Content demand status, which is either hot or cool, is applicable only when using the hot content load balancing feature. For a given hot content subset, content demand status is cool from the time the cache rule is implemented until the number of requests for the subset exceeds the hot_threshold during a hit_period. At this point, content demand status for the subset becomes hot, and requests for any item in the subset are load balanced to the hot_pool. Content demand status remains hot until the number of requests for the subset falls below the cool_threshold during a hit_period, at which point the content demand status becomes cool. The BIG-IP then directs requests for any item in the subset to the appropriate server in the cache_pool until such time as the subset becomes hot again. For the configuration shown in Figure 12.1, on page 12-1, you would create a rule to cache all content having the file extension .html or .gif.

To create a cache statement rule using the Configuration utility

1. In the navigation pane, click Rules.
   The Rules screen opens.
2. Click the Add button.
   The Add Rule screen opens.
3. In the Add Rule screen, type the cache statement.
   To cache all content having either the file extension .html or .gif, you would type:
   
   

   rule cache_rule { cache ( http_uri ends_with "html" or http_uri ends_with "gif" ) { origin_pool origin_server cache_pool cache_servers hot_pool hot_cache_servers } }

4. Click the Add button.

To create a cache statement rule from the command line

Given the configuration shown in Figure 12.1, on page 12-1, to cache all content having the file extension .html or .gif, you would use the bigpipe command:

b 'rule cache_rule { \

cache ( http_uri ends_with "html" or http_uri ends_with "gif" ) \

{ origin_pool origin_server \

cache_pool cache_servers \

hot_pool hot_cache_servers } }'

Creating a virtual server

Now that you have created pools and a cache statement rule to determine how the BIG-IP will distribute traffic in the configuration, you need to create a virtual server to use this rule and these pools. For this virtual server, use the host name or IP address that Internet clients use to request content from your site. To create the configuration shown in Figure 12.1:

• Add a virtual server with address 10.10.10.4 and port 80 (this means the virtual server will accept traffic for the HTTP service only).
• Add the rule cache_rule.

To create a virtual server using the Configuration utility

1. In the navigation pane, click Virtual Servers.
2. Click the Add button.
   The Add Virtual Server screen opens.
3. For each virtual server, enter the virtual server address and pool name. For additional information about configuring a virtual server, click the Help button.

To create a virtual server from the command line

Use the bigpipe virtual command to configure the virtual server to use the pool that contains the outside addresses of the firewalls:

b virtual 10.10.10.4:80 use rule cache_rule

Configuring for intelligent cache population

Your cache rule routes requests to either the origin server or to the appropriate cache server.

When the cache rule directs a request from a user to the origin server, the BIG-IP translates the destination of the request to the origin server and translates the source of the request to the translated address and port of the associated Secure Network Address Translation (SNAT) connection. This ensures that the request reaches the origin server and that the origin server responds to the BIG-IP and not directly to the user.

When the cache rule directs a request from a user to the cache server, the cache will not contain the requested content if either it is the first time a cache has received a request for the content or the content has expired. In this case, the cache initiates a miss request (that is, a request resulting from a request for content a cache does not have) for this content to the origin server specified in the configuration of the cache or to another cache server. If you want to allow intelligent cache population, you should configure the cache with its origin server set to be the virtual server on the BIG-IP, so that the cache sends miss requests to the internal shared interface of the BIG-IP. The BIG-IP translates the destination of the request, and sends the request to either the origin server or another cache server that already has the requested content.

To ensure that the origin server or cache server responds to the BIG-IP rather than to the original cache server that generated the miss request, the BIG-IP also translates the source of the miss request to the translated address and port of the associated SNAT connection.

In order to enable these scenarios, you must:

• Create a SNAT for each cache server.
• Identify the origin server node as remote.

Configuring a SNAT

The SNAT translates the address of a packet from the cache server to the address you specify. For more information about SNATs, see SNATs in the BIG-IP Reference Guide, Chapter 4, Configuring the High-Level Network. To create the configuration shown in Figure 12.1, on page 12-1, use the translation address 10.10.10.5.

To configure a SNAT mapping using the Configuration utility

1. In the navigation pane, click SNATs.
   The SNATs screen opens.
2. Click the Add button.
   The Add SNAT screen opens.
3. In the Add SNAT screen, configure the attributes required for the SNAT you want to add. For additional information about configuring a pool, click the Help button.

To configure a SNAT mapping from the command line

To configure a SNAT mapping from the command line, type:

b snat map 10.10.20.4 10.10.20.5 10.10.20.6 to 10.10.10.5

Configuring a SNAT automap for bounceback

You must now configure a second SNAT mapping, in this case using SNAT automap, so that when requests are directed to the origin server, the server will reply through the BIG-IP and not directly to the client. (If this were to happen, the next request would then go directly to the origin server, removing the BIG-IP from the loop.)

To configure a SNAT automap from the command line

Configure the existing SNAT address 10.10.10.5 on the external interface as a self address.

b self 10.10.10.5 vlan external snat automap enable

Enable SNAT auto-mapping on the external VLAN:

b vlan external snat automap enable

Additional configuration options

Whenever a BIG-IP is configured, you have a number of options:

• You have the option in all configurations to configure a BIG-IP redundant system for fail-over. Refer to Chapter 6, Configuring a Redundant System, in the BIG-IP Reference Guide.
• All configurations have health monitoring options. Refer to Health Monitors in Chapter 4, Configuring the High-Level Network, in the BIG-IP Reference Guide.
• When you create a pool, there is an option to set up persistence and a choice of load balancing methods. Refer to Pools in the Chapter 4, Configuring the High-Level Network, in the BIG-IP Reference Guide.