OS tuning considerations
Due to each customer's unique H/W configuration and workload, each customer's environment may generate different system bottleneck and require tailored OS tunings. NetBackup technical support published known OS tunings that help with system performance and will continue to do so in the future as new tunings are discovered. Beyond those already published tech notes, Veritas recommends running with the default OS setting first. If, as more loads are added, throughput continues improve and the four major system resource usage increase proportionally, then most likely the system is tuned properly. You will know when the system requires tuning, if there is unexpected hardware resource bottleneck or throughput does not improve while plenty of h/w resources is still available. See the flow chart in Chapter 5 for the process of performance issue troubleshooting.
Following are symptoms and remedies for some common system resource bottlenecks that NBU servers may encounter:
Table:
Resource | Symptom | Recommendaiton |
|---|---|---|
Memory | Heavy swapping |
|
Network | Poor network traffic load balancing |
|
Client connection failure due to exceeding max socket connections | Example: Set net.core.somaxconn=1024 (default=128) on Red Hat OS | |
CPU | CPU idle dropped to below 15% | Try reducing the load. |
I/O | Poor backup performance during Read/write mixed workload | If you use Veritas Infoscale for MSDP storage pool management, the poor performance could be caused by the memory map (mmap) lock contention. For NetBackup versions 8.2, 8.3, 8.3.0.1, 9.0, and 9.0.0.1, contact Veritas support for the EEB that fixed the mmap lock issue. |
Poor I/O performance due to LUN sharing | Avoid LUN sharing by relocating the second volume to a dedicated LUN if one exist | |
Open files | Maximum open files exhausted | Try doubling the current setting |