Quantum 6-01376-05 User Manual

Page of 32
  StorNext File System Tuning
The Metadata Controller System
StorNext File System Tuning Guide
14
The 
cvcp
 utility is a higher performance alternative to commands such as 
cp
 and 
tar
. The 
cvcp
 utility achieves high performance by using threads, 
large I/O buffers, preallocation, stripe alignment, DMA I/O transfer, and 
Bulk Create. Also, the 
cvcp
 utility uses the SNFS External API for 
preallocation and stripe alignment. In the directory-to-directory copy 
mode (for example, 
cvcp source_dir destination_dir
,) cvcp conditionally 
uses the 
Bulk Create
 API to provide a dramatic small file copy 
performance boost. However, it will not use 
Bulk Create 
in some 
scenarios, such as non-root invocation, managed file systems, quotas, or 
Windows security. Hopefully, these limitations will be removed in a 
future release. When 
Bulk Create
 is utilized, it significantly boosts 
performance by reducing the number of metadata operations issued. For 
example, up to 20 files can be created all with a single metadata 
operation. For more information, see the 
cvcp
 man page. 
The 
cvmkfile
 utility provides a command line tool to utilize valuable 
SNFS performance features. These features include preallocation, stripe 
alignment, and affinities. See the 
cvmkfile
 man page.
The 
Lmdd
 utility is very useful to measure raw LUN performance as well 
as varied I/O transfer sizes.
The 
cvdbset
 utility has a special “Perf” trace flag that is very useful to 
analyze I/O performance. For example: 
cvdbset perf
Then, you can use 
cvdb -g 
to collect trace information such as this:
PERF: Device Write 41 MB/s IOs 2 exts 1 offs 0x0 len 0x400000 mics 95589 ino 
0x5
PERF: VFS Write EofDmaAlgn 41 MB/s offs 0x0 len 0x400000 mics 95618 ino 
0x5
The “PERF: Device” trace shows throughput measured for the device I/
O. It also shows the number of I/Os into which it was broken, and the 
number of extents (sequence of consecutive filesystem blocks).
The “PERF: VFS” trace shows throughput measured for the read or write 
system call and significant aspects of the I/O, including:
• Dma:  DMA
• Buf:  Buffered
• Eof: File extended
• Algn: Well-formed DMA I/O
• Shr: File is shared by another client
• Rt: File is real time