Versions Compared

Key

  • This line was added.
  • This line was removed.
  • Formatting was changed.

...

Parallel writes is a configurable feature relating to the number of web service requests across an instance of DryvIQ, on a given node, that will operate in parallel. However, it is very important to note that increasing the number of parallel writes does not always equal faster/better. There is a long list of concepts that have to be taken into account. Consult your Professional Consultative Services Representative representative or Customer Support for assistance understanding hw this setting can impact your jobs and configuration. Also, if you are running DryvIQ in an autoscale auto scale environment, this does not apply as scaling will occur as configured.

...

A job in DryvIQ does not use a fixed amount of memory. Memory usage for individual jobs will vary based on a number of factors, the most significant one is the number of files and how those files are distributed (all in one folder, throughout sub-folders, etc.). To avoid excessive memory usage related to how content is distributed, DryvIQ recommends preserving the system default for Directory Item Limits. The main factors for memory usage for a DryvIQ node will be number of concurrent jobs plus Parallel Writes Per Job for each job and the memory impact of the specific jobs.

...

If memory issues occur due to increasing the Directory Item Limit or Parallel Writes Per Job, there is no other mitigation other than reducing the number of current jobs or breaking up the source content in to multiple jobs. DryvIQ will keep using memory until it runs out (it will not self-limit) and it will eventually reach the environment max. Reaching an environment max may result in a non-graceful termination of DryvIQ that could result in jobs re-transferring files, permissions or metadata.  In the case of larger jobs being stopped in this manner, they will enter recovery mode, continue to use all the memory, then get stopped again, in a loop causing a loss of throughput.

...