What It Means


Everyone's IO Profile is unique to their hardware/software system and the applications that are running.  Each Capture is comprised of many different IO streams, but is usually dominated by a handful that comprise the majority of IOs.  IO Streams in application space will be changed as they traverse the hw/sw stack and will be different when presented to the storage.  You should observe how many and what percent the dominant IO streams occur in your capture.

More Specifically

1. Generally state the character of your capture (i.e. predominantly RND and SEQ smaller block Rs and Ws or perhaps predominantly larger block SEQ Rs.)
How many IO streams comprise at least 80% of the IOs that occured?
What are the main characteristics?  - RND or SEQ, block size and RW mix?
Can you state these major IO streams generally for your capture activity? 

2. Compare this characterization to commonly accepted definitions of your workload activity (i.e. streaming VOD is usually larger block Rs.)
Does your capture support or refute the commonly accepted definition?
Did you run only the assumed workload? (i.e. were any other applications running while you took your capture?)
Did you run multiple captures to see if it is repeatable?

3. How does your capture activity change when you change variables?
Is it the same when you use similar, but different, sw?  (i.e.Youtube vs Amazon.)
Is it different when you use different hw/sw? (i.e. Is it the same on a Mac vs a PC? Chrome vs Safari?)
Does it change if you use different storage?  (i.e. Is it the same on a HDD vs an SSD?)
Is your capture repeatable if taken at different levels of storage fill?  (i.e. Is it the same on an SSD that is 50% full vs one that is 80% full or one that is brand new?  Is your application running in a virtualized server or stand alone PC?)

4. How does your capture compare to captures taken by others?
Is there a similar workload capture posted on the IPF database?  Note: you must be an IPF Portal member to view the IPF database.
Are there captures posted that are taken on different hw/sw?

5. How does the workload characterization compare to published SSD specifications?
Does your HDD or SSD specification list a workload similar to the one you captured?
If not, does the workload on the specification bear any resemblance to your capture?