Communications Server

Network Administration Guide

Chapter 6. Planning for HPR

Communications Server supports high performance routing (HPR) over Enterprise Extender (IP), synchronous data link control (SDLC), LAN, WAN, channel, Multi-Path Channel (MPC), and X.25 connections.

HPR automatic network routing (ANR) minimizes the storage and processing requirements in intermediate nodes, which is an improvement over APPN intermediate session routing (ISR) for high-speed networks with low error rates.

Communications Server HPR support includes support for rapid transport protocol (RTP) and automatic network routing (ANR).

HPR Link Characteristics

To provide the benefits listed above, HPR assumes that data links provide low rates of lost frames. HPR recovers from conditions that cause lost frames, but if the conditions persist for long periods, HPR assumes that network congestion is too high and HPR throughput is lowered. Achieving good HPR throughput on a given link involves proper configuration of the sending node, the facilities that constitute the link between the two nodes, and the receiving node.
Note:You might want to set ERP values in cases where LAN line facilities have high frame loss.

Sending Node

Typical adapters used with HPR will send frames onto the line with low rates of failure. For Ethernets, HPR behavior will be degraded if the LAN is heavily loaded with many collisions (for example, if the load is more than about 30% to 40% of the nominal Ethernet speed). For frame relay, facilities should be configured so that individual connections can burst to the access line rate without causing congestion conditions in the frame-relay network.

Line Facilities

LAN facilities can have substantial hidden complexity, ranging from large numbers of stations sharing a single LAN segment to bridges between LAN segments, perhaps with hidden lower speed links, to simulated LAN segments that operate over complex subnets. The low frame loss assumption must be met. For example, WAN links between split bridges must use facilities with low error rates or must compensate by using link level error recovery between the bridges. As another example, conditions that cause bridge buffers to overrun (broadcast storms or large bursts of broadcast frames) must be prevented.

Receiving Node

Typical adapters used with HPR will receive frames from the line with low rates of failure. Some adapters, however, will fail to receive frames when consecutive frames arrive too quickly. The best HPR performance is realized when the receiving adapter can handle its traffic without missing inbound frames.

If your network has a high error rate, systems administrators may notice lower than expected throughput on HPR connections.

HPR Considerations

The HPR features are achieved by including additional information in frame headers. If you do not need the HPR features, the additional overhead may reduce your network performance. The HPR features are an advantage whenever there are intermediate nodes or alternate routes. However, if you require high performance between two nodes connected by a single link without non-disruptive recovery during a link failure, you should disable HPR when you configure that link.

HPR implements priority queuing. In general, interactive traffic passes through the network faster than lower priority batch traffic. Adapters that queue frames do so without regard to priority and may interfere with HPR's priority queueing. You may need to reduce the adapter's queue depth to prevent low priority traffic from delaying high priority traffic. (See the adapter documentation for instructions on setting the adapter queue depth.)

Many PCs now have advanced power management features that will suspend power to devices (like disk drives) and reduce the clock rate on a CPU significantly to save power if the keyboard or mouse is not being used. On a workstation, this is a good feature, but on a server platform, this can occur while it is routing data from one adapter to another (and no disk activity is occurring). For Communications Server, disable advanced power management and hardware power management by entering the BIOS configuration during boot up (usually pressing F1 while initializing hardware) and disable these options.

HPR will use more of the LAN adapter memory due to increased performance and queueing. Many shared RAM adapters default to using 8 kilobytes of shared RAM. It is required that the shared RAM size is 16 kilobytes. To adjust this configuration, refer to the adapter configuration manual. This setting is located in either your machine BIOS, or on the adapter switch settings.
Note:If you do not increase the shared RAM to 16 kilobytes, network performance is degraded.

Do not adjust the shared RAM value above 16 kilobytes, the optimal setting for the adapter memory.

If HPR links are failing under heavy traffic, you can set your receive timers to a higher value. This allows more time for the links to receive an acknowledgement for sent frames.

For SDLC links, this is the Primary Receive Timer. The receive timer can be set on either the DLC or a logical link statement.

To calculate the minimum receive timer value in tenths of seconds, use the following formula:

[(send_window_count * max_i_field_size * 8) /actual line speed]* 10

Considerations for Tuning Parameters on LAN DLCs

You can specify eight parameters to affect how quickly Communications Server handles recovery and failure of LAN links. You might need to adjust these parameters to achieve the best results if you are using slow links, HPR, or backup host connections.

You should consider the following parameters when defining a LAN device in SNA Node Configuration:

Under the Advanced tab:

Test retry interval
The test retry interval specifies the time between attempts to find the adjacent link station on the local area network (LAN). The number of attempts made are based on the value specified for the test retry limit.

The default value is 8.

Test retry limit
The test retry limit specifies the maximum number of attempts to find the adjacent link station on the local area network (LAN) without receiving an acknowledgment in the time set by the value for the test retry interval.

The default value is 5.

Maximum PIU size
The maximum path information unit (PIU) size is the maximum number of bytes that are contained in the data buffer used by SNA sessions for this link. If the value of the maximum PIU size exceeds the frame size supported by your local device driver, the value will be reduced to match the frame size.

The default value is 65535.

Under the Performance tab:

Idle timeout
The idle timeout specifies the time that the LAN device driver waits for a frame to be received before declaring the link to be inoperative.

The default value is 30 seconds.

Acknowledgment delay
The acknowledgment delay specifies the time that the LAN device withholds a response to a received frame in order to allow more frames to be received and acknowledged with the same Request Ready (RR).

The default value is 100 milliseconds.

Poll response timeout

The poll response timeout specifies the time that the LAN device waits for a response to a frame sent with the POLL bit set.

The default value is 8 000 milliseconds.

Anticipated outstanding transmits
The anticipated outstanding transmits specifies the maximum number of frames the LAN device queues to a link station before sending a Receive Not Ready (RNR) to the adjacent link station.

The default value is 16 frames.

Receive buffer count
The receive buffer count specifies the number of buffers that are reserved in memory to hold data received from the host until it can be processed. Each buffer is the size of the PIU.

The default value is 32 buffers.

HPR Path Switch Processing

HPR provides a non-disruptive path switch to a different connection if one connection fails. A retry of the failed connection is attempted before a path switch occurs. An explanation of the parameters involved in determining whether a connection has failed and the time to initiate a path switch follows.

When the value specified for Idle timeout is reached, a keepalive frame is sent. The default for Idle timeout is 30 seconds.

The value specified for Poll response timeout determines the time that the device waits for a response to the keepalive frame. The default for Poll response timeout is eight seconds (8 000 milliseconds). A keepalive frame is retried 10 times. This retry count can not be changed.

If there is no response to the keepalive frame, TEST commands are sent to the partner system. The value specified for Test retry interval determines the time between TEST command attempts, and the value specified for Test retry limit determines the maximum number of TEST command attempts. The default for Test retry interval is 8 seconds, and the default for Test retry limit is five attempts.

The following calculation indicates the time required to determine whether a connection has failed and initiate a path switch:

(Idle timeout) + (Poll response timeout * 10 retries) + 
           (Test retry interval * Test retry limit) + 1

Using the defaults for the parameters, the time required to determine a connection failure and initiate a path switch can be up to 151 seconds.

(30 seconds) + (8 seconds * 10 retries) + 
           (8 seconds * 5 attempts) + 1 = 151 seconds

If you are experiencing long path switch times (or failures) with HPR when a connection fails, you can reduce the time by decreasing the values for the parameters in the calculation.

[ Top of Page | Previous Page | Next Page | Table of Contents | Index ]