This section describes capacity planning for the SteelHead. This section includes the following topics:

This section describes the characteristics of the optimization resources available for the SteelHeads. These resources are the primary determining factors for supported WAN capacity, maximum number of optimized TCP connections, and RiOS data store capacity. For example, the amount of hard drive space and RAM determine how large the RiOS data store can be.

This section includes the following topics:

For more information about individual model specs, go to http://www.riverbed.com/products-solutions/products/.

Each SteelHead model has a maximum number of optimized TCP connections. The larger the SteelHead, the more CPU, memory, and disk resources are available, increasing the amount of supportable connections. This is one of the primary considerations you use for sizing branch offices. Typically, Riverbed recommends a guideline of 5-10 connections per user. Sizing for data center SteelHeads must take in account all optimized connections coming into the data center. For planning purposes, use the high end numbers for these types of connections. Large amounts of active connections (connections that are actively transmitting data), such as HTTP, have more impact on the SteelHead resources.

To view the connection history use the show stats connection <timeframe> command.

You can see the average and maximum number of connections for the time frame entered. These numbers are useful to determine if the current SteelHead is properly sized for the number of connections. You can also use this command to review when adding addition users or applications that can increase the number of optimized connections.

WAN capacity limit is the amount of optimized traffic that is sent outbound from a SteelHead. This is also commonly used for sizing branch office SteelHeads. For data center SteelHeads, the WAN capacity is a recommendation on how much typical data throughput a SteelHead can process. Model 1050 and below are limited to the rated outbound capacity (outbound after optimization). No rate limit or hard restriction is applied to throughput for a data center SteelHead (2050 and larger), though excessive amounts of certain types of traffic can strain the resources available on the SteelHead.

RiOS data store size is the amount of disk space, in GB, available for SDR use. The data center SteelHeads use RAID 10 or FTS 7050 for disk redundancy and optimal performance.

Take disk performance into account for high-end data center deployments. Certain types of data put more load on the disk systems and can be monitored if performance issues are suspected. High throughput data replication deployments typically use dedicated SteelHeads.

Monitor the disk systems with the following OID.

Use the following Data Store Disk Load report in the Management Console to monitor disk performance.

If the Disk Load report is showing 80-90% for a sustained amount of time, or multiple times a day that coincide with periods of lower performance and average RiOS data store cost greater than 5000, the disk load might be impacting overall performance.

This section describes admission control and This section includes the following topics:

Admission control prevents the SteelHead from processing traffic when overloaded. It also controls the connection count limits. Admission control stops the interception of connections for optimization but still allows the connections to pass through without optimization. Admission control is in one of two states:

Each model contains connection limits to limit the total number of connections that is accepted into the system. The connection limits have rising and falling thresholds. The rising threshold is the cutoff limit. While the system is in flowing state, if the connection count rises above this threshold, admission control moves to the paused state. The falling threshold is the enable limit. While the system is in the paused state, until the connection count falls below this threshold, admission control keeps the system in paused state.

Some leeway is given for connection limits before admission control is triggered. The minimum number of additional allowed connections before entering admission control on any model is 10. For example, a 250L with a rating of 30 connection does not enter admission control until it passes 40 connections. The SteelHead does not exit admission control until the number of connections falls back below the rated limit. Using the same example, a 250L entering admission control at 40 connections does not start optimizing new connections again until it is back down below 30 connections (rated limit).

The SteelHead sends out a SNMP alert, called Admission Control Error, to the SNMP host that you define. It alerts you that the licensed optimization limit is reached. You can purchase a bigger SteelHead that can take all the optimized TCP connections; or you can limit the type of traffic to be optimized with in-path rules configuration, and ensure maximum optimization benefits are limited only to most critical traffic or specific traffic that is hogging the WAN bandwidth.

Riverbed recommends polling the number of optimized connections periodically, so you can take proactive steps before all the optimized TCP connections are consumed. The following table shows the SNMP MIB for the number of optimized connections.

Each SteelHead model contains memory limits to limit the total amount of memory that is used. The memory limits have rising and falling thresholds. The rising threshold is the cutoff limit. While the system is in flowing state, if the memory usage rises above this threshold, admission control moves to the paused state. The falling threshold is the enable limit. When the system is in the paused state, until memory usage falls below this threshold, admission control keeps the system in paused state.

For additional information about admission control, see the Riverbed Knowledge Base article Understanding Admission Control at https://supportkb.riverbed.com/support/index?page=content&id=s15140.