Quantum 6-01376-05 User Manual

  StorNext File System Tuning

The Metadata Controller System

StorNext File System Tuning Guide

8

Some metadata operations such as file creation can be CPU intensive, and 

benefit from increased CPU power. The MDC platform is important in 

these scenarios because lower clock- speed CPUs such as Sparc and Mips 

degrade performance.
Other operations can benefit greatly from increased memory, such as 

directory traversal. SNFS provides three config file settings that can be 

used to realize performance gains from increased memory: 

, 

, and 

. 

However, it is critical that the MDC system have enough physical 

memory available to ensure that the FSM process doesn’t get swapped 

out. Otherwise, severe performance degradation and system instability 

can result.

0

The following FSM configuration file settings are explained in greater 

detail in the 

 page. For a sample FSM configuration file, 

see 

The examples in the following sections are excerpted from the sample 

configuration file from 

0

Splitting apart data, metadata, and journal into separate stripe groups is 

usually the most important performance tactic. The 

, 

, and 

 (e.g., write) operations are very sensitive to I/O latency of the 

journal stripe group. Configuring a separate stripe group for journal 

greatly benefits the speed of these operations because disk seek latency is 

minimized. However, if 

, 

, and 

 performance aren't 

critical, it is okay to share a stripe group for both metadata and journal, 

but be sure to set the exclusive property on the stripe group so it doesn't 

get allocated for data as well. It is recommended that you assign only a 

single LUN for each journal or metadata stripe group. Multiple metadata 

stripe groups can be utilized to increase metadata I/O throughput 

through concurrency. RAID1 mirroring is optimal for metadata and 

journal storage. Utilizing the write-back caching feature of the RAID 

system (as described previously) is critical to optimizing performance of 

the journal and metadata stripe groups.