This question was raised by a client – whilst doing an investigation into the underlying VMware and SAN infrastructure of a couple of systems for a customer where users were reporting poor application performance – as being a potential avenue for analysis.
The application in question reads from an SQL database, then writes the read file to local disk as a temporary file. When Process Monitor (available in the Sysinternals Suite and freely downloadable from Microsoft) is run against the file system to see the transacation taking place, a ReadFile length of 4096, and WriteFile length of 1024 is recorded.
Fig. 1 Process Monitor ReadFile Length 4096
Fig. 2 Process Monitor WriteFile Length 1024
When looking at this, I could find little information on the Net to explain this behaviour, so the following is a hypothesis.
Assuming that the ReadFile length mentioned is 4096 bytes, and the WriteFile length is 1024 bytes, these numbers could come from:
Default Network Packet Size on SQL Server = 4096 bytes
Fig. 3 Network Packet Size setting taken from SQL Server Management Studio: Database Properties > Advanced
NTFS Default Bytes Per FileRecord Segment = 1024 bytes (at least where the Bytes Per Cluster is 4096)
Fig. 4 Bytes Per FileRecord Segment from Windows Command Prompt:\> fsutil fsinfo ntfsinfo C:
Please feel free to comment – especially if my hypothesis is wrong (which is part of the reason for posting.)
Post a Comment