I am trying to assess the performance of my scientific software using Process Explorer, with the files both locally on our blade server (Dell VRTx, Windows Server 2012) and over the network from another Windows machine (Windows 7 machine). What I have found is that the performance drops off when comparing the local and the network operations, which makes sense. What I don't understand is that when running the software using data on the networked machine share, all operations start using 4k blocks instead of the 1.2k blocks I saw when the data was local to the machine.
Is there something within TCP or SMB that is controlling the block size for read/write operations from computer to NAS? Would this have something to do with TCP buffering? Data is below, thanks in advance for any assistance.
Software run with data on local disk
- Average Write Block size: 3894kB
- Average Read Block size: 1162kB
- Max IO Demand Rate: 535.24MB/sec
- Max CPU Demand Rate: 24677.33 Mil Cycles/sec
Software run with data on networked machine share
- Average Write Block size: 3900
- Average Read Block size: 3900
- Max IO Demand Rate: 141.88MB/sec
- Max CPU Demand Rate: 12000.14 Mil Cycles/sec
Thanks in advance.
