Guidelines for setting DELAY

To specify the amount of time that elapses before VTAM® or NCP sends buffered data, use the DELAY operand on NCP and VTAM definition statements. The DELAY operand on the NCP BUILD definition statement controls how long the NCP buffers data; it is the elapsed time between the receipt of the first inbound message and the presentation of an attention interrupt to the host. This capability can affect your strategy for increasing coattailing use of the channel.

The DELAY operand that controls how long VTAM buffers data can be specified on the following statements:

PCCU definition statement for NCPs not defined in a channel-attachment major node
GROUP, LINE, or PU definition statements for NCPs defined in a channel-attachment major node
PU definition statement for local (channel-attached) SNA controllers

The following three examples illustrate the use of the DELAY operand.

Figure 1. Effect of DELAY time on coattailing - example 1

Example of Effect of Delay time on coattailing.

Figure 2. Effect of DELAY time on coattailing - example 2

Figure 3. Effect of DELAY time on coattailing - example 3

For NCP, the DELAY operand allows time for more than one message to arrive in the NCP before an attention is presented to the host.

The following events cause PIUs on the queue to be sent before the DELAY time elapses:

VTAM receives an attention interrupt from NCP.
VTAM or NCP reaches the MAXBFRU limit.
VTAM sends virtual route pacing responses.
VTAM sends transmission priority two traffic.
VTAM sends APPN network priority or highest priority traffic (to a local SNA PU).

The default value for the DELAY operand is 0 (for example, no coattailing). Take the default under the following conditions:

You have a low rate of transactions that use the channel connection, and response time for interactive transactions is important.
You have a high rate of transactions that use the channel connection, and response time for interactive transactions is important.
You have predominantly unidirectional traffic (file transfer) with little other activity in the opposite direction.
You are benchmarking single or few sessions of bulk data traffic (file transfers, backups, and so on).
You have channel traffic with frequent bursts of SENDs (such as from an application program that makes inquiries of the other host in rapid succession).

To analyze inbound coattailing for channel-attached SNA communication or cluster controllers, compare the number of times that NCP signals VTAM that it has data to send (ATTN tuning statistic) to the number of times that VTAM issues a channel READ program to read data (CHRD or RCH tuning statistic). A less accurate method is to divide the number of inbound PIUs transferred by NCP (IPIU tuning statistic) by the number of read channel programs issued by VTAM (CHRD or RCH). This calculation gives you the average number of PIUs read from the channel-attached device with each channel program, which is an indicator of coattailing.

Note: For a 3274 to clear its buffers after a data transfer operation, the 3274 signals to VTAM that it has more data to send even though there is no data. This causes the RDATN tuning statistic to equal the CHRD or RCH tuning statistic, and because RDATN is included in the ATTN tuning statistic, the ATTN value is normally double the CHRD or RCH value.

To analyze outbound coattailing for channel-attached SNA communication or cluster controllers, divide the number of outbound PIUs (OPIU tuning statistic) by the number of WRITE channel programs VTAM issues (CHWR or WCH tuning statistic). This calculation gives you the average number of PIUs transferred by VTAM with each channel write program, which is a rough estimate of the effectiveness of coattailing.

Note: The DELAY specification for a 3174 is a customization option. However, even if you set the DELAY customization option for a 3174 that is performing a token-ring gateway function, the option is ignored. The DELAY operand in VTAM should also be set to 0 for a 3174 that is a token-ring gateway channel-attached cluster controller.

Specifying a nonzero attention DELAY value to activate attention DELAY does not automatically ensure coattailing. To ensure coattailing, all of the following must be true:

The installation must have a transaction rate high enough to allow more than one message to accumulate in the DELAY time period in the NCP or the host.
The MAXBFRU value must be large enough to allow more than one message to be sent to the host.
The host processing speed has an effect on coattailing; the faster host processors reduce coattailing slightly. An attention interrupt occurs when MAXBFRU is reached or when the DELAY time period expires.

Begin by setting DELAY equal to or greater than 0.2. A 0.2-second delay has little effect on response time, but if the traffic speed is approximately 15 transactions per second or greater, coattailing occurs.