Configuring TCP Parameters

You can control TCP behavior in a switch by configuring the TCP parameters that are described in this section.

Note     When FCIP is sent over a WAN link, the default TCP settings may not be appropriate. In such cases, we recommend that you tune the FCIP WAN link by modifying the TCP parameters (specifically bandwidth, round-trip times, and CWM burst size).

This section includes the following topics:

Configuring Minimum Retransmit Timeout

You can control the minimum amount of time TCP waits before retransmitting. By default, this value is 200 milliseconds (msec).

Configuring Keepalive Timeout

You can configure the interval that the TCP connection uses to verify that the FCIP link is functioning. This ensures that an FCIP link failure is detected quickly even when there is no traffic.

You can configure the first interval during which the connection is idle (the default is 60 seconds). When the connection is idle for the configured interval, eight keepalive probes are sent at 1-second intervals. If no response is received for these eight probes and the connection remains idle throughout, that FCIP link is automatically closed.

Note     Only the first interval (during which the connection is idle) can be changed.

Configuring Maximum Retransmissions

You can specify the maximum number of times a packet is retransmitted before TCP decides to close the connection.

Configuring Path MTUs

Path MTU (PMTU) is the minimum MTU on the IP network between the two endpoints of the FCIP link. PMTU discovery is a mechanism by which TCP learns of the PMTU dynamically and adjusts the maximum TCP segment accordingly (RFC 1191).

By default, PMTU discovery is enabled on all switches with a timeout of 3600 seconds. If TCP reduces the size of the maximum segment because of PMTU change, the reset-timeout specifies the time after which TCP tries the original MTU.

Configuring Selective Acknowledgments

TCP may experience poor performance when multiple packets are lost within one window. With the limited information available from cumulative acknowledgments, a TCP sender can only learn about a single lost packet per round trip. A selective acknowledgment (SACK) mechanism helps overcome the limitations of multiple lost packets during a TCP transmission.

The receiving TCP sends back SACK advertisements to the sender. The sender can then retransmit only the missing data segments. By default, SACK is enabled on Cisco MDS 9000 Family switches.

Configuring Window Management

The optimal TCP window size is automatically calculated using the maximum bandwidth parameter, the minimum available bandwidth parameter, and the dynamically measured round-trip time (RTT).

Note     The configured round-trip-time parameter determines the window scaling factor of the TCP connection. This parameter is only an approximation. The measured RTT value overrides the round trip time parameter for window management. If the configured round-trip-time is too small compared to the measured RTT, then the link may not be fully utilized due to the window scaling factor being too small.

The min-available-bandwidth parameter and the measured RTT together determine the threshold below which TCP aggressively maintains a window size sufficient to transmit at minimum available bandwidth.

The max-bandwidth-mbps parameter and the measured RTT together determine the maximum window size.

Note     Set the maximum bandwidth to match the worst-case bandwidth available on the physical link, considering other traffic that might be going across this link (for example, other FCIP tunnels, WAN limitations). Maximum bandwidth should be the total bandwidth minus all other traffic going across that link.

Configuring Monitoring Congestion

By enabling the congestion window monitoring (CWM) parameter, you allow TCP to monitor congestion after each idle period. The CWM parameter also determines the maximum burst size allowed after an idle period. By default, this parameter is enabled and the default burst size is 50 KB.

The interaction of bandwidth parameters and CWM and the resulting TCP behavior is outlined as follows:

The software uses standard TCP rules to increase the window beyond the one required to maintain the min-available-bandwidth to reach the max-bandwidth.

Note     The default burst size is 50 KB.

Tip     We recommend that this feature remain enabled to realize optimal performance. Increasing the CWM burst size can result in more packet drops in the IP network, impacting TCP performance. Only if the IP network has sufficient buffering, try increasing the CWM burst size beyond the default to achieve lower transmit latency.

Configuring Estimating Maximum Jitter

Jitter is defined as a variation in the delay of received packets. At the sending side, packets are sent in a continuous stream with the packets spaced evenly apart. Due to network congestion, improper queuing, or configuration errors, this steady stream can become lumpy, or the delay between each packet can vary instead of remaining constant.

You can configure the maximum estimated jitter in microseconds by the packet sender. The estimated variation should not include network queuing delay. By default, this parameter is enabled in Cisco MDS switches when IPS modules or MPS-14/2 modules are present.

The default value is 1000 microseconds for FCIP interfaces.

Configuring Buffer Size

You can define the required additional buffering—beyond the normal send window size —that TCP allows before flow controlling the switch’s egress path for the FCIP interface. The default FCIP buffer size is 0 KB.

Note     Use the default if the FCIP traffic is passing through a high throughput WAN link. If you have a mismatch in speed between the Fibre Channel link and the WAN link, then time stamp errors occur in the DMA bridge. In such a situation, you can avoid time stamp errors by increasing the buffer size.


Copyright 2010-2013, Cisco Systems, Inc. All rights reserved.