For testing caching solutions, normal IOMeter is ineffective since the data is 100% random and there is no hot data to cache. IOMeter should have settings (or a version?) to allow various non-random patterns, such as a normal distribution.
Does anyone else have the demand for this feature? Do you mean something like a gaussian curve
Or is it sufficient and realistic enough to add some kind of locality or skew (% of IOs go to % of space)?
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I have prototyped changes for precisely this purpose, using the concept of locality of reference (LOR) for reads and writes. I would definitely like to pursue this conversation. LOR is more representative of real world applications, and will be important for testing many of the newly emerging storage architectures.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I have prototyped changes for precisely this purpose, using the concept of locality of reference (LOR) for reads and writes. I would definitely like to pursue this conversation. LOR is more representative of real world applications, and will be important for testing many of the newly emerging storage architectures.
[feature-requests:#26] Normal distribution data patterns for IOMeter
Status: open
Created: Fri Nov 16, 2012 07:12 PM UTC by Anonymous
Last Updated: Mon Jul 15, 2013 09:16 AM UTC
Owner: nobody
For testing caching solutions, normal IOMeter is ineffective since the data is 100% random and there is no hot data to cache. IOMeter should have settings (or a version?) to allow various non-random patterns, such as a normal distribution.
Does anyone else have the demand for this feature? Do you mean something like a gaussian curve
Or is it sufficient and realistic enough to add some kind of locality or skew (% of IOs go to % of space)?
I have prototyped changes for precisely this purpose, using the concept of locality of reference (LOR) for reads and writes. I would definitely like to pursue this conversation. LOR is more representative of real world applications, and will be important for testing many of the newly emerging storage architectures.
Hi Brian,
Thanks! We are open to all contributions. Please send me your implementation and we will review it.
Thanks,
Ved
On Thursday, February 6, 2014 7:55 PM, Brian Martin kestrelv@users.sf.net wrote:
I have prototyped changes for precisely this purpose, using the concept of locality of reference (LOR) for reads and writes. I would definitely like to pursue this conversation. LOR is more representative of real world applications, and will be important for testing many of the newly emerging storage architectures.
Related
Feature Requests: #26