The Traffic Management Microkernel (TMM) processes all load-balanced traffic on the BIG-IP system. TMM runs as a real-time user process within the BIG-IP operating system (TMOS). CPU and memory resources are explicitly provisioned in the BIG-IP configuration.
Understanding BIG-IP CPU usage
The following factors influence the manner in which TMM uses the CPU:
- The number of processors installed in the BIG-IP system
- The BIG-IP version
- The modules for which the BIG-IP system is licensed
CPU utilization on single CPU, single core systems
CPU resources are explicitly provisioned in the BIG-IP configuration. When TMM is idle or processing low volumes of traffic, TMM yields idle cycles to other processes.
CPU utilization on multi-CPU / multi-core systems
Prior to BIG-IP 11.5.0, each logical CPU core is assigned a separate TMM instance, and each core processes both data plane (TMM-specific) tasks and control plane (non-TMM-specific) tasks.
Beginning in BIG-IP 11.5.0, data plane tasks and control plane tasks use separate logical cores on systems with Intel Hyper-Threading Technology (HT Technology) CPUs. Even-numbered logical cores (hyperthreads) are allocated to TMM, while odd numbered cores are available for other processes.
Using the tmsh utility to view TMM CPU usage
- Log in to the TMOS Shell (tmsh) by typing the following command:
tmsh
- To display TMM CPU utilization and other statistical information for TMM instances, type the following tmsh command:
show /sys tmm-info
For example, the following tmsh command is showing CPU usage for TMM 0.0 (Output truncated):Sys::TMM: 0.0 -------------------------- CPU Usage Ratio (%) Last 5 Seconds 3 Last 1 Minute 3 Last 5 Minutes 2
- Average over all TMM CPUs (all even CPUs)
- Average over ‘all odd CPUs except the last one’ (The reason for leaving out the last CPU is due to an analysis plane that was spiking the last CPU numbers.)
The higher of these values are presented as the overall system CPU usage.
Understanding BIG-IP Memory usage
When administering a BIG-IP system, it is important to understand how the system allocates memory. In general, BIG-IP memory usage falls into the following categories:
- Traffic Management Microkernel (TMM) memory usage
- Linux host memory usage
- Swap usage
TMM runs as a real-time user process within the Linux host operating system. The BIG-IP system statically assigns memory resources to TMM and potentially to other module-related processes, depending on module provisioning. The remaining memory is available for all other Linux host processes.
The BIG-IP system creates swap usage space during software installation on disk. Swap space is available to the Linux kernel.
TMM memory usage
The BIG-IP data plane includes one or more TMM processes to manage traffic on the BIG-IP system. The BIG-IP system statically assigns memory resources to TMM.
The following information summarizes TMM memory:
- The BIG-IP system assigns a dedicated pool of memory to each TMM process.
- TMM memory is not available for the Linux kernel to reassign to other host processes. The system never considers TMM memory as available.
- TMM memory cannot be swapped to disk.
- The TMM memory management subsystem allocates and clears memory pages in the following manner:
- TMM allocates static memory to hash tables (for example, the connection flow table).
- TMM dynamically allocates memory pages for temporary objects (for example, persistence records and buffered connection data).
- Memory sweepers periodically reap unused memory as needed from TMM objects.
- When possible, TMM caches dynamic allocations to improve performance when new objects require the same allocations.
Linux memory usage
The system may allocate remaining memory to other processes on the Linux host and kernel threads.
The following information summarizes Linux host memory usage:
- Linux allocates most available memory to buffers and disk caching, which gives the appearance of high memory usage but allows the system to run more efficiently.
- Linux utilities, such as top and free, may report that only a small amount of memory is free. This is normal behavior; cached memory can be reclaimed quickly if a program needs memory.
- To see memory used by buffers and disk caching, view the -/+ buffers/cache row where top and free report these memory structures. Add these values to the reported amount of free memory to estimate the total amount of physical memory the processes are not currently using.
- The Linux kernel sometimes copies memory pages to swap. This is known as swapping memory.
Swap memory usage
The following information summarizes swap memory usage:
- It is normal for a Linux system, including the BIG-IP system, to use a small amount of swap. The Linux kernel sometimes prefers to swap idle processes memory to disk so that more physical memory is available for more active processes, buffers, and caches.
- Physical memory is much faster than swap, and prioritizing buffers and caches allows the kernel to optimize performance of disk-heavy processes such as databases.
- A higher percentage of swap use is normal when provisioned modules make heavy use of the disk.
- Excessive swap usage may be a sign that the system is experiencing memory pressure. You should investigate in the following cases:
- The system uses a very high percentage of swap memory.
- The percentage of swap memory usage increases over time.
Understanding BIG-IP memory statistics
You can view BIG-IP memory statistics using BIG-IP utilities or Linux command line utilities. It is normal for Linux utilities, such as top and free, to report a small amount of free memory. This expected behavior occurs due to Linux disk caching. F5 recommends that you use the Configuration utility or the TMOS Shell (tmsh) to view memory statistics on the BIG-IP system.
You can view BIG-IP memory statistics, including TMM memory usage, other (Linux) memory usage, swap usage, and memory allocated to TMM hash tables and cache objects. To do so, use the following utilities:
- tmsh show /sys memory
- Configuration utility: Statistics > Module Statistics > Memory
Memory statistics (BIG-IP 10.x - 11.5.4)
In BIG-IP 10.x - 11.5.4, the Configuration utility tmsh report memory allocated to buffers and caches as used memory. As a result, it may appear that the host system is using all available memory. The system reports memory statistics in the following ways:
- System Memory
- Host Total: The amount of memory available to Linux or non-TMM processes.
- Host Used: The amount of memory in use by Linux or non-TMM processes.
- TMM Total: The amount of memory available to TMM processes.
- TMM Used: The amount of memory in use by TMM processes for traffic management.
- Subsystem memory/memory pool name
- Indicates the name and memory utilization of TMM hash tables and cache objects.
Source:f5.com
No comments:
Post a Comment