0

Imagine running a test where you have a network-attached storage device, and there are two running processes (on different machines), reading and writing the same file on the NAS. Process 'W' does dedicated writes, and process 'R' does dedicated reads.

What is the correct terminology for the total time it takes a single byte written by 'W' to be made available for reading by 'R'? Is this throughput, latency or something else? This would factor in all protocol and network related latency.

Edit: Actual problem is discussed below:

We have a NAS device on our network, and a requirement that we read a growing file being written by one device (the 'W' process described above), on a second device (the 'R' process). We need to read very close to the tail of the file, and periodically we receive I/O errors ("Broken pipe"). We are trying to determine what is causing the broken pipe and tune the NAS accordingly.

Improve this question
13
  • What protocol do the computers use to access the NAS?
    – David Schwartz
    Feb 15, 2012 at 16:58
  • The writer process connects from Windows via CIFS, and the reader process connects on a Mac (Snow Leopard) via SMB (also CIFS, I suppose).
    – Dan
    Feb 15, 2012 at 17:12
  • 2
    I will suggest that two programs that need to communicate in a tightly-coupled fashion should not do it via a file on remote storage.
    – mfinni
    Feb 15, 2012 at 17:27
  • @Dan: Does the reader hold file read locks when it reads/stats and the writer hold file write locks when it writes? How does the reader decide when to read/stat?
    – David Schwartz
    Feb 15, 2012 at 17:38
  • @mfinni I don't disagree, however, I have to work within our client's requirements.
    – Dan
    Feb 15, 2012 at 17:40

2 Answers 2

Reset to default
0

Depending on where in the two-way chain of events you place this term, it could be "commit rate" or "commit time".

Improve this answer
0

I'm not sure you understand the question, is this homework, if so please read our FAQ, we're not here for that kind of thing.

Anyway the answer is 'who knows' - because it depends on how long after the write it's committed on the filer and when the file is unlocked and its metadata is updated to reflect the change, only then will the reading app realise its cached copy has been updated so will read the updated data.

There's no set-time that's universal.

Improve this answer
4
  • Homework ? on servefault ? oh noes!
    – adaptr
    Feb 15, 2012 at 16:26
  • This isn't homework, I am trying to troubleshoot what seems to be a disk performance problem but I don't know how to configure my test tools to get the correct metrics.
    – Dan
    Feb 15, 2012 at 16:29
  • maybe discussing the actual problem with actual details might be more productive than a hypothetical
    – Chopper3
    Feb 15, 2012 at 16:32
  • Ok, edited as requested.
    – Dan
    Feb 15, 2012 at 16:54

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .