White Papers

BP1015 Scalability & Deployment Best Practices for the Dell EqualLogic FS7500 NAS System 9
Table 3 CIFS Simulation: File System Operation Distribution
Operation Type
Percent
READ_ANDX
20%
WRITE_ANDX
9%
OTHER
7
71%
Distribution: READ Only
Less than 4KB
60%
4KB - 32KB
20%
32KB - 256KB
20%
Distribution: WRITE Only
Less than 4KB
67%
4KB - 32KB
18%
32KB - 256KB
15%
3.5 Performance criteria
The use case considered for simulations in this paper is end-user access of file shares. The response
time expectation for end-user file share interaction from a client-side application is typically in the 1-2
seconds range. At approximately 1 second response time, end-users start to notice a delay. But, a 1
second delay should not cause end-users to divert their attention from the application.
8
At 0.1 second
(100ms) users typically notice no delay and perceive the application response to be instantaneous. For
our file share simulations we chose the response time requirement for file operations to be under
100ms.
9
In fact, the actual measured response times during our simulations were much lower than
this arbitrary limit.
Response time is measured from test clients for operations completed on the NAS file shares. Note
that enterprise server applications using a NAS as backend storage will need much faster average I/O
response times than 100ms.
3.6 Simulation methodology
The simulation workload was applied or requested from test clients on the NAS system in increasing
order of file I/O operations per second. The achieved file operations per second was measured and
reported. If the achieved operations per second were not close to the requested operations per
second, then the system was deemed to be saturated at an earlier requested load point. Also, average
response time of the operations was measured across test clients to make sure they stayed within the
established criteria. The achieved operations per second close to the saturation level was considered
the optimal operations per second achievable for the configuration.
7
The OTHER category includes primarily NFS meta-data operations such as ECHO, LOCKING_ANDX,
NT_TRANSACT_QUERY_SECURITY_DESC, TRANS2_QUERY_FS_INFORMATION,
TRANS2_QUERY_FILE_INFORMATION, TRANS2_QUERY_PATH_INFORMATION,
TRANS2_SET_FILE_INFORMATION, TRANS2_FIND_FIRST2, NT_CREATE_ANDX, CLOSE, and FLUSH.
8
http://www.useit.com/papers/responsetime.html
9
Miller, R. B. (1968). Response time in man-computer conversational transactions -
www.mattson.com.au/robert/files/Miller1968.pdf