IT Operations Management (ITOM)
cancel

Five questions to ask before you upgrade to the 64-bit HPE Operations Agent 12.00 on HP-UX

Five questions to ask before you upgrade to the 64-bit HPE Operations Agent 12.00 on HP-UX

GirishMatti

HPE Operations Agent (OA) version 12.00 on HP-UX is available in native 64-bits as well as in the 32-bits avatar. The 32-bits compiled OA on a HP-UX server runs in an emulation mode on the 64-bit HP-UX platforms (11.23 and 11.31). One of the most common questions I get is if users should upgrade to this new version or if they should remain with their current version.

To help solve this dilemma for yourself, here’s a quick test to decide if you need the native 64-bit HPE Operations Agent 12.00 on HP-UX 11.23 and HP-UX 11.31 or if you should continue with the 32-bit HPE Operations Agent.

Get your pencils ready—it’s test time!


Question 1: Are you seeing incorrect or partial performance metric values?

Answer: On a 64-bit HP-UX 11.23 or 11.31, if you see incorrect, partial or corrupted performance metric values, then move to the 64-bit agent.


Question 2: Does your server have a huge number of CPUs and other hardware resources or a huge amount of performance data is collected?

Answer: If yes, then the 64-bit agent should suit you.

Note: Some brief background on the reasons: On HP-UX the performance metrics are collected by the process ‘midaemon’ which interacts with the kernel instrumentation to collect these metrics. The kernel instrumentation shares the raw data (called traces) via shared memory, which is used by midaemon to derive the performance metrics. The amount/volume/speed at which the traces are served by the kernel instrumentation depends on the size of the system. The 32-bit HPE Operations agent has a limitation of addressing only up to 2GB of shared memory. There is an option of enabling the executable option called Exec-Magic which allows an additional 1GB of shared memory addressing.

The 32-bit architecture works well on smaller- and medium-sized servers where the requirement for addressing the shared memory is mostly within the addressable limits. On high-end servers such as Superdome having 100s of CPUs, 3GB of shared memory is not enough. Even if the shared memory is available on the system, midaemon will not be able to address memory larger than 3GB. Therefore the data collection tools such as glance/scopeux/perfd may give incorrect /partial data and core dump sometimes due to invalid memory access.

Now, back to the test…


Question 3: Do you need PRM (Process Resource Manager) related metrics?

Answer: If so, then wait for the next release of the HPE Operations Agent, which is scheduled for later this year. This future release will bring back support for PRM metrics.

With HPE OA 12.00, these metrics would show up as NA or 0 on all the direct clients of HPE Operations agent that is Glance, Scopeux, Perfd, XGlance, CODA and the clients of CODA.

4) Do you need Network metrics related to X.25 and ATM?

Answer: If yes, then stick with the 32 bits agent as the 64 bits agent cannot collect these metrics. These metrics will show up as NA or 0 on all the direct clients of HPE Operations agent that is Glance, Scopeux, Perfd, XGlance, CODA and the clients of CODA.


5) What happens to my existing performance data?

Answer: Upon upgrading from any previous version of HPE Operations agent to 12.00 (which is a native 64 bit agent), the coda, scopeux database files and DSI log files cannot be migrated. The existing log files created by the 32-bit agent will be backed up in the /var/opt/perf/Backup_32_bit_PA directory.


So what is your choice: 32-bit or 64-bit HPE Operations Agent?


Also, please read the HPE Operations Agent release notes for more details and other limitations.


Discover-2016-LV-blog-banner.jpg

  • operations bridge
About the Author

GirishMatti