Type
  • Best Practice
Description
  • This article explains the recommended configuration for optimal performance of the Hardware Sentry KM.
Additional Keywords
  • Hardware Sentry KM for PATROL, System Requirements

Related Topics

Knowledge Base

Performance and Stability Best Practices for Hardware Sentry KM

KB1270 - May 27, 2020

System Requirements

Sentry Software recommends having at minimum, per instance of PATROL Agent and Hardware Sentry KM:

  • 2 CPU cores
  • 4 GB of RAM

This may need to be increased, depending on the number of other KMs loaded/running at the same time, additional PATROL Agents running on the same system, and the size of the environment.

Increasing the Java Collection Hub Maximum Heap Size

If your system is configured as above, but the KM runs out of memory without reaching the physical limit, then you may need to configure the Collection Hub Heap Size Maximum, as detailed in KB 1240.

For Hardware Sentry KM, /SENTRY/HARDWARE/collectionHubHeapSizeMax should be set to 1024(MB) or more.

Workload

For optimal performance and stability, we recommend not exceeding 8,000 instances per PATROL Agent, which is typically the equivalent of about 80 servers or 16 medium-sized storage systems.

The number of instances created will depend on the number of physical components of the system (eg, Disks, Network Interfaces, etc.), and will vary. For servers, the average number of instances is about 100, with each instance having an average of 4 or 5 parameters. For storage systems, the number of instances will be much higher.

Some very large storage systems (EMC Symmetrix, Hitachi G-series, etc.) may have thousands of disks/volumes. For example, in our testing, one NetApp FAS6040 had 265 Instances and 777 Parameters. Thus, it is important to keep in mind the instance count when planning the number of required PATROL Agents.

The current instances/parameters count can be obtained by either of the following methods:

  1. Executing the PATROL SID Report in a TrueSight Query Window:

  2. Executing this PSL script in a TrueSight Query Window:

    foreach k(get_vars("/", "subnodes")){s+=get("/".k."/numInstances");}print(s);

Additional Factors

There are other factors to consider, which will affect the performance and resource usage of Hardware Sentry KM. These include:

  • Protocols in use (SNMP, WBEM, WMI, HTTP, SSH, etc.). SNMP is generally the least resource-intensive/fastest, while SSH/Command-Line is the most resource-intensive/slowest.
  • Network reliability.
  • Responsiveness of the remote systems to monitor.

An excessive number of timeouts/retries may lead to unpredictable memory/CPU consumption by the PATROL Agent or KM’s Collection Hub, which will impact the monitoring.

Recommendations

  • PATROL Agent version should be at least 11.3.02i.
  • It is recommended to check the events generated for the ProtocolStatus and the connector's Status parameters often, as they are a good indicator of the connection's consistency.
  • The recommended discovery/collect intervals for a large environment are a 1 hour discovery interval, and 5 or more minutes collect interval.