Guest

Cisco BPX/IGX/IPX WAN Software

Real-Time Throttling Mechanisms in IPX/IGX Switch Software

Cisco - Real-Time Throttling Mechanisms in IPX/IGX Switch Software

Document ID: 10800

Updated: Oct 04, 2005

   Print

Introduction

This document explains real-time throttling mechanisms in Cisco/StrataCom IPX switch software. Real-time Throttles describe what a throttle is and why throttling is needed. Specific Throttles describe specific mechanisms, how real-time is improved by their fine tuning, trade-offs, and release-specific default configurations.

Prerequisites

Requirements

There are no specific requirements for this document.

Components Used

This document is not restricted to specific software and hardware versions.

Conventions

Refer to Cisco Technical Tips Conventions for more information on document conventions.

Real-time Throttles

A real-time throttle is a mechanism that regulates the amount of work performed by the PCC card. A throttle permits the user to reduce or smooth work over time and frees the PCC card to perform other tasks. Sometimes limiting the work or tasks performed by a specific mechanism has few adverse affects and is preferable in systems where real time is limited or close to critical levels. This document explains such throttles and can help you evaluate which tasks are most appropriately abbreviated in a specific network environment for the sake of increasing available real time.

Specific Throttles

Comm-Fail Test Throttle I

Description

The Communication Failure test identifies trunks that cannot successfully pass a test pattern in the IPX system and consists of sending network test messages periodically over each trunk within a node. Throttling is accomplished with the Comm Fail Interval and Comm Fail Minimum configuration parameters. The Comm Fail Interval parameter defines the cycle time for all trunks within a node. Therefore, a cycle time of 5 minutes for a node with ten lines results in one Comm-Fail test execution every 30 seconds. The Comm Fail Minimum parameter defines the minimum period and overrides the former parameter when the calculated interval is less.

The Communication Failure test is modified in Release 6.3 and is scheduled to run on each trunk independently. In addition, the Comm Fail Minimum parameter is displaced by the Comm Fail Multiplier parameter, both of which are used to configure a modified scheduling algorithm. Trunks without a communication failure in progress are scheduled for testing every Interval * Multiplier milliseconds. Trunks with a communication failure in progress are scheduled for testing every Interval milliseconds. Increasing interval and multiplier configuration values spreads work over time.

Trade-offs

For networks where Comm-Breaks are configured as informational, trade-offs are minimal. A truly bad line takes longer to diagnose. For networks where Comm-Breaks result in connection derouting, an undiagnosed faulty trunk can result in declaration of Comm-Break before declaration of Comm-Fail. The former failure indicates an inability to communicate with a remote node for an unknown reason and results in connection derouting while the latter failure results in immediate connection rerouting (around the identified bad trunk).

Configuration

Command Parameter Name 6.1 6.2 6.3 Units
cnfnodeparm Comm Fail Interval 5000 5000 10000 msecs
Comm Fail Minimum 700 700 - msecs
Comm Fail Multiplier - - 3 -

Comm-Fail Test Throttle II

Description

The Communication Failure test pattern (the payload that is inserted into test packets) is configurable and can be shortened to reduce work. Default configuration is three packets, which you can reduce to two packets or one packet.

Trade-offs

Shorter test patterns are not as robust. The Trade-offs that Comm-Fail Test Throttle I describes also apply here.

Configuration

Command Parameter Name 6.1 6.2 6.3 Packets Units
cnfcftst None 30 30 30 3 bytes
None 16 16 16 2 bytes
None 2 2 2 1 bytes

Comm-Fail Test Throttle III

Description

The Communication Failure test can be delayed from running immediately after switchovers or rebuilds with the Comm Fail Delay configuration parameter. Operation of this mechanism permits lines to exit comm-fail status but not enter comm-fail status until this timer expires. This mechanism has proven valuable in dealing with peak periods of CPU usage typically encountered after switchovers or rebuilds where comm-fails are falsely declared solely because of a congested system. False declaration adds to system congestion. It initiates connection rerouting around the presumed failed trunk. This throttle permits the Comm-Fail test to avoid a known vulnerable period and resume when real-time usage returns to normal levels.

Trade-offs

A bad trunk is not detected until the timer expires.

Configuration

Command Parameter Name 6.1 6.2 6.3 Units
cnfnodeparm Comm Fail Delay - - 60 Seconds

Comm-Break Test Throttle I

Description

Sometimes a node within a network cannot communicate with another node. Nodes in this state are deemed unreachable to each other and continually seek to reestablish communication by means of the Communication Break Test. Once this test passes, a flurry of real-time intensive database updates are exchanged between the nodes, and processing returns to normal. Increasing the time between tests is an effective means of staging the work required to process database updates and is the rationale behind the Comm-Break Test Throttle. Throttling is controlled with the Comm-Break Test Delay and the Comm-Break Test Offset configuration parameters. Both parameters are used to configure the interval of time between Comm-Break test initiation. The former parameter is used when many nodes are unreachable and the latter parameter is used when few nodes are unreachable. You should maintain the default ratio between the Comm-Break Test Delay and the Comm-Break Test Offset that appears in the configuration table, below. To implement a different two-tiered time-out feature, consult Cisco/StrataCom Software Engineering.

Trade-offs

No adverse side affects exist in networks where Comm-Break declaration is informational. The restoration of derouted connections to service takes longer in networks where Comm-Break declaration results in connection derouting.

Configuration

Command Parameter Name 6.1 6.2 6.3 Units
cnfnodeparm Comm-Break Test Delay 10000 30000 30000 msecs
Comm-Break Test Offset 10 10 10 -

Comm-Break Test Throttle II

Description

The Communication Break test pattern length is configurable in 1-, 4-, or 16-packet messages and can be shortened to reduce work.

Trade-offs

Shorter test patterns are not as robust as longer patterns. A line can be falsely declared okay, and later be declared bad. This side effect is minimal in networks with informational Comm-Breaks.

Configuration

Command Parameter Name 6.1 6.2 6.3
cnfcbtst None See screen See screen -

Updates Delay Throttle

Description

Immediately after PCC switchover, databases are periodically exchanged or updated per node under the control of the switched node. The extension of the internode update time is the rationale behind the Updates Delay Throttle. Throttling is controlled by the Update Initial Delay and the Update Per-Node Delay configuration parameters. The Update Initial Delay parameter designates a wait period before it updates the first node and the Update Per-Node Delay parameter specifies remaining wait periods.

Trade-offs

Few. Distributed databases take longer to reconcile.

Configuration

Command Parameter Name 6.1 6.2 6.3 Units
cnfnodeparm Update Initial Delay 1000 5000 5000 msecs
Update Per-Node Delay 30000 30000 30000 msecs

TXR Signaling Throttle I

Description

The network communication protocol can be adapted to process messages directed at a single node in parallel. Most StrataCom networks employ single-threaded node-to-node network communications, however, networks that employ TXR signaling may require multiple threading. Throttling is performed with the NW Sliding Window Size configuration parameter. Default configuration assumes no TXR signaling.

Trade-offs

Dial-pulsed signaling can become distorted when this number is set too low.

Configuration

Command Parameter Name 6.1 6.2 6.3 Units
cnfnodeparm NW Sliding Window Size 1 1 1 Messages processed in parallel

TXR Signaling Throttle II (Release 6.3)

Description

Time-out periods for the network communication handler processing can be fine-tuned in systems with TXR signaling. This timer controls the nonevent initiation of the network handling function and is used to increase network processing in systems with TXR signaling. Increasing the frequency (decreasing the time-out) of Network Handler functioning is done to interrogate for incoming signaling message completion. In the absence of this time-out the Network Handler initiates only when outgoing messages are queued. Default configuration assumes no TXR signaling.

Trade-offs

None in systems without TXR signaling.

Configuration

Command Parameter Name 6.1 6.2 6.3 Units
cnfnodeparm NW Hdlr Timer - - 50 msecs

Adaptive Voice Throttle

Description

The Adaptive Voice feature can be tailored to specific network environments for improved real-time. This feature continually seeks to translate unused trunk bandwidth into higher grade voice connections. This process is referred to as normalization.

The frequency of normalization attempts and the number of voice connections normalized per attempt constitute the Adaptive Voice Throttle. This throttle includes the Normalization Interval, Max Number to Normalize, and the Settling Interval configuration parameters. The Normalization Interval parameter is the time between attempts to normalize connections. The Max Number to Normalize parameter is the maximum number of connections per normalization attempt that can enter the normalized mode. The Settling Interval is the time after a trunk failure during which no normalization attempts are made. This gives all nodes a chance to reroute without competition by the adaptive voice feature for trunk bandwidth.

Trade-offs

None. Normalization is simply a marginal improvement over an already high-quality voice connection.

Configuration

Command Parameter Name 6.1 6.2 6.3 Units
cnfcmparm Normalization Interval 2 2 2 Minutes
Max Number to Normalize 5 5 5 Connections
Settling Interval 4 4 4 Minutes

Reroute Bundling Throttle

Description

The maximum number of connections that can be routed in a single attempt is configured with the Maximum Routing Bundle configuration parameter. Routing many small connection bundles has the effect of smoothing routing work over a longer period of time. Better load balancing is also a benefit with small bundles.

Trade-offs

Connections in excess of the bundle size that requires routing are delayed. This can be okay with some connection types. Thought should be given about the type of traffic routed in a network and the network protocols supported.

Configuration

Command Parameter Name 6.1 6.2 6.3 Units
cnfcmparm Maximum Routing Bundle 24 24 24 Connections

Reroute Throttle

Description

Rerouting can be throttled with the Reroute Timer and Reset Timer on Line Fail parameters. The former parameter specifies a time-out that starts when a connection is routed, and prohibits the connection from attaining routing candidacy until it has expired. The latter parameter is an on/off switch; it overrides the Reroute Timer in the event that a line fail is the impetus for connection derouting. This throttle might be considered to slow rerouting in networks experiencing intermittent trunk failures on preferred routes (for example, networks experiencing adverse weather conditions such as electrical storms).

Note: Connections are no longer derouted due to endpoint failures (removed cards, FAILED cards, and so forth.) in Release 6.2 or later, and consequently the Reroute Timer is not initialized for such failures.

Trade-offs

Connections can remain derouted longer than desired.

Configuration

Command Parameter Name 6.1 6.2 6.3 Units
cnfcmparm Reroute timer 0 0 0 Seconds
Reset Timer on Line Fail Yes Yes Yes Yes/No

Down/Up Connection Throttle

Description

Connections are downed and upped in configurable bundle sizes with an interbundle time-out period. The Max Down/Up Per Pass and Down/Up Timer configuration parameters specify the number of connections per bundle and the time that must elapse before the next bundle is upped/downed. Small bundles and long inter-bundle periods can be implemented to smooth work over time in networks that employ this feature frequently.

Trade-offs

None.

Configuration

Command Parameter Name 6.1 6.2 6.3 Units
cnfcmparm Max Down/Up Per Pass 50 50 50 Connections
Down/Up Timer 30000 30000 30000 msecs

Statistics Collection Throttle

Description

Statistics collection is throttled with the Statistical Sample, Frp Port Sampling, and Conn Sampling on/off switches. These three switches control CBUS messaging activity and are used to disable statistics collection during peak periods of real-time usage, such as software upgrades.

Trade-offs

Statistics are lost while sampling is disabled.

Note: StrataView+ should be restarted after these parameters are turned back on.

Configuration

Command Parameter Name 6.1 6.2 6.3 Units
on1 / off1 Conn Sampling On On On On/off switch
on2 / off2 Statistical Sample On On On On/off switch
FRP Port Sampling On On On On/off switch

Connection Priority Bumping Throttle

Description

The System-Wide Priority Bumping feature continually seeks to route connections based on a class-of-service (COS) scheme in which each connection is assigned a numerical value (COS) that establishes its priority relative to other connections. Connections with high COS can bump those with low COS and reoccupy the desired route. This feature is throttled with four configuration parameters, which include the Number of connections processed per pass, Time between idle processing, Time to next COS processing, and Time between failed candidate retries.

The first parameter specifies the number of candidate connections per pass where a pass consists of several sub-steps for each COS. The sub-steps are:

  1. Candidate connections with the selected COS are identified.

  2. Lower COS connections are bumped.

  3. Higher COS connections are routed.

  4. The feature waits before it starts the next COS.

The second and third parameters specify the length of wait periods between passes and COS processing respectively. The final parameter specifies the wait period which must elapse before a connection that has failed to secure a route can again attain candidacy to bump. Extension of time-out periods and the reduction of the number of connections per pass smoothes bumping work over time.

Trade-offs

The desired state of routed and non-routed connections takes longer to attain.

Configuration

Command Parameter Name 6.1 6.2 6.3 Units
cnfcmparm Number of connections... 2000 2000 2000 Connections
Time between idle proc.. 60 60 60 Connections
Time to next COS processing 45 45 45 Connections
Time between failed cand.... 10 10 10 Connections

Diagnostic Throttle

Description

Background Diagnostics and Selftest execution can be throttled to run less frequently if disabled. Tests are configurable separately and by card type.

Trade-offs

A failure that can be diagnosed can take longer to identify or may not be identified.

Configuration

Command Parameter Name 6.1 6.2 6.3
cnftstparm Enable Enabled Enabled Enabled
Time-out See screen See screen See screen

Real Time Profiler Throttle

Description

Real Time is continually monitored by the Real Time Profiler, which periodically snapshots salient real-time indicators and saves them in memory for display. This period, which is identified as the Interval Time, can be increased to reduce PCC work load.

Trade-offs

Information is presented in fewer data points - loss of resolution.

Configuration

Command Parameter Name 6.1 6.2 6.3 Units
cnfprfparm Interval Time 20 20 20 Seconds

Standby PCC Database Update Throttle (Release 6.3+ only)

Description

Standby PCC database updates are throttled with the Standby Update Timer and the Stby Updts Per Pass configurable parameters. The Standby Update Timer parameter controls the time between each database update pass and the Stby Updts Per Pass parameter controls the number of databases updated per pass. In order to smooth the work load, extend the time-out period and decrease the number of databases transferred per pass.

Trade-offs

An unexpected switchover or rebuild can result in loss of configurable information (for example, connections, trunks, configurable parameters). Use of this throttle widens a window of vulnerability. However, this risk has presented few real problems in the past.

Configuration

Command Parameter Name 6.1 6.2 6.3 Units
cnfnodeparm Standby Update Timer - - 10 Seconds
Stby Updts Per Pass - - 30 Databases

Download Throttle

Description

Downloading is throttled with the Rmt Blk Freq, Rmt Blk Size, Lcl Blk Freq, and Lcl Blk Size configuration parameters. Remote parameters refer to internode downloading and the local parameters refer to Active to Standby PCC downloading and StrataView+ to Active PCC downloading.

Increase the block frequency (read time-out) and decrease the block size to spread work over a longer period of time.

Trade-offs

Downloading takes longer to complete.

Configuration

Command Parameter Name 6.1 6.2 6.3 Units
cnfdlparm Rmt Blk Freq 100 100 100 msecs
Rmt Blk Size 0x400 0x400 0x400 bytes
Lcl Blk Freq 100 100 100 msecs
Lcl Blk Size 400 400 400 bytes

Global Connection Rerouting Throttle (Release 6.3+)

Description

Global connection rerouting is throttled with the Gateway ID Timer and the GLCON Alloc Timer configurable parameters. These parameters throttle two of the three steps necessary to route global connections. Global connection routing steps include Gateway Identification, GLCON allocation, and routing.

Gateway ID Timer expiration initiates a search to identify gateway nodes for up to the Maximum Routing Bundle (see cnfcmparm) number of derouted global connections per pass. Gateway nodes to enter and exit the junction domain are identified for each connection at a connections master node. Extend this time-out to smooth routing work over a longer period of time.

GLCON allocation timer expiration initiates allocation of GLCON data structures at gateway nodes identified in the previous step. Extend this time-out to smooth routing work over a longer period of time.

Trade-offs

Global connections take longer to route.

Configuration

Command Parameter Name 6.1 6.2 6.3 Units
cnfnodeparm Gateway ID Timer - - 30 Seconds
GLCON Alloc Timer - - 30 Seconds

Related Information

Updated: Oct 04, 2005
Document ID: 10800