Quantum 6-01376-05 User Manual

  StorNext File System Tuning

The Underlying Storage System

StorNext File System Tuning Guide

4

it severely degrades typical scenarios. Therefore, it is unsuitable for most 

environments.

0

Configuration settings such as RAID level, segment size, and stripe size 

are very important and cannot be changed after put into production, so it 

is critical to determine appropriate settings during initial configuration.
The best RAID level to use for high I/O throughput is usually RAID5. 

The stripe size is determined by the product of the number of disks in the 

RAID group and the segment size. For example, a 4+1 RAID5 group with 

64K segment size results in a 256K stripe size. The stripe size is a very 

critical factor for write performance because I/Os smaller than the stripe 

size may incur a read/modify/write penalty. It is best to configure 

RAID5 settings with no more than 512K stripe size to avoid the read/

modify/write penalty. The read/modify/write penalty is most 

noticeable in the absence of “write-back” caching being performed by the 

RAID controller.
The RAID stripe size configuration should typically match the 

SNFS 

 configuration setting when multiple LUNs are utilized in a 

stripe group. However, in some cases it might be optimal to configure the 

SNFS StripeBreadth

 as a multiple of the RAID stripe size, such as when 

the RAID stripe size is small but the user's I/O sizes are very large. 

However, this will be suboptimal for small I/O performance, so may not 

be suitable for general purpose usage.

 is the best RAID level for metadata and journal storage 

because it is most optimal for very small I/O sizes. It is also very 

important to allocate entire physical disks for the Metadata and Journal 

LUNs in order to avoid bandwidth contention with other I/O traffic. 

Metadata and Journal storage requires very high IOPS rates (low latency) 

for optimal performance, so contention can severely impact IOPS (and 

latency) and thus overall performance. If Journal I/O exceeds 1ms 

average latency, you will observe significant performance degradation.
It can be useful to use a tool such as 

 to help determine the storage 

system performance characteristics and choose optimal settings. For 

example, varying the stripe size and running lmdd with a range of I/O 

sizes might be useful to determine an optimal stripe size multiple to 

configure the SNFS 

.