A Simple Example
April 1, 2014 § 2 Comments
Taking an 8Gb/s fibre channel port, I’m going to use Queuing Theory, and some simplifying assumptions, to determine at what utilization latency becomes significant enough to worry about. Whether, and to what extent, these assumptions are justified is a discussion for another time. The aim is to make the analysis easy while still maintaining sufficient realism to be useful.
Let’s assume we are throwing data about and all our frames are carrying a full payload, giving a frame length of 2,112 bytes. With 8b/10b encoding, this ends up being 21,120 bits. 21,120 bits / 8Gb/s gives us an expected service time of 2.5 microseconds. To make this easier to deal with, we’ll assume the port has infinite buffers, that each frame arrives randomly and independently (a Poisson process) and that the service times are similarly distributed. None of these are entirely realistic assumptions, but they make it a lot easier to construct an analytical model.
With the above assumptions, we can use the formula for the responce time of an M/M/1 queue:
Where 
If we plot this we find that the port appears to cope very well under high load:
at 95% utilization, the port has only just hit 50 microseconds of response time. Given that the best case response time of a storage array is, in most cases, in the low hundreds of microseconds it’s clear that, in this example, queuing in the buffers on fibre channel ports can be discarded as a source of latency in all but the most extreme cases.
In future posts, I will explore ways in which this analysis can be extended and cases in which fibre channel network performance can impact storage performance.
ascknd 
My relatives always say that I am killing my time here at net, except I know I am getting experience all the time by reading such good posts.
Your style is very unique in comparison to other people I’ve read stuff from.
Thanks for posting when you have the opportunity, Guess I will
just book mark this blog.