Memory consideration for IBM Spectrum Scale cache usage

This topic describes how the Spectrum Scale metadata for open files will increase in relation to the maxFilesToCache value.

In Figure 1, a value of 100k for maxFilesToCache already needs around 230 MiB memory for metadata compared to 45MiB for 20k.
Figure 1. Spectrum Scale maxFilesToCache metadata for open files
This graphic shows for IBM Spectrum Scale, the Metadata used for open files related to maxFilesToCache.

The open file metadata size can be checked using the mmdiag –stats command.

This example shows how to query the open file count metadata for mFTC, which has been set to 100k:

# mmdiag --stats
=== mmdiag: stats ===
Global resources:
  OpenFile counts: total created 104795 (in use 100000, free 4795)
    using 232812K memory
    cached 100000, currently open 2+3, cache limit 100000 (min 10, max 100000), 
    eff limit 100000
    stats: ins 20869218 rem 20769218 creat 1548983 destr 1444188 free 10496248 
    reuse 10491453 steals 10390822 (clean 9391515, dirty 999307)
A rule of thumb: The total amount of memory Spectrum Scale uses to cache file data and metadata is estimated by adding together:
  1. The pagepool size.
  2. The amount of memory required to hold inodes and control data structures (maxFilesToCache × 3 KiB).
  3. The memory for the stat cache (maxStatCache × 400® bytes).