Re: [Iometer-devel] Access histograms and libaio

[Will M. <wjm...@gm...>]

On 9/10/2010 10:56 PM, Vedran Degoricija wrote:
> Hi Will,
> These features have been considered and are on the TODO list. However, any help is
> appreciated.
> Iometer currently uses AIO. AFAIK, it should be the same AIO as FIO uses. However,
> FIO has other interfaces such as the kernel IO one, and this is what Iometer
> lacks. Do you have any experience with this?
Sorry, no experience in that area yet. We will undoubtedly be taking a closer look 
at this and what it entails.

> As for histograms, I had an exchange with some of the developers and saving off
> all per-IO information might accure too much data and/or too much overhead for
> high IOPS capable devices. There are other ways to do this in a way which will
> have minimal impact to the system.
This was one of my concerns, as a single run could easily run into the millions of 
IOPS. Our "solution" to that thus far would likely be to throw RAM at the system, 
but even that would accrue overhead.

Were those exchanges here on the mailing list, and if so how far back? I'm all 
ears for low impact solutions, so if someone has a good idea I'm willing to look 
at implementing one.

> Ved
>
> ----------------------------------------------------------------------------------
> *From:* Will Marone <wjm...@gm...>
> *To:* iom...@li...
> *Sent:* Thu, September 9, 2010 5:30:55 PM
> *Subject:* [Iometer-devel] Access histograms and libaio
>
> All,
>
> A co-worker and I are looking to add a pair of features to IOMeter. Specifically,
> we're looking at adding libaio support for Linux (similar to fio) and per-access
> timestamps that could be used to generate a histogram. Right now we're at the
> 1000m view of IOMeter as a whole, so some clarification on the questions:
>
> Currently, I understand that IOMeter uses "aio" for accesses, however I'm not sure
> if that is the same "aio" that fio uses, or if it is a different one. If anyone
> could clarify this, it would be greatly appreciated.
>
> As for the histogram, the current idea is (for the time being) to locally cache IO
> timestamps for each IO request for a given run then dump to a disk on the machine
> running dynamo, with the long term goal of reporting these back to the host and
> including them in the regular reports. Is there any existing work in this area?
>
> Will
>
> ------------------------------------------------------------------------------
> Automate Storage Tiering Simply
> Optimize IT performance and efficiency through flexible, powerful,
> automated storage tiering capabilities. View this brief to learn how
> you can reduce costs and improve performance.
> http://p.sf.net/sfu/dell-sfdev2dev
> _______________________________________________
> Iometer-devel mailing list
> Iom...@li... <mailto:Iom...@li...>
> https://lists.sourceforge.net/lists/listinfo/iometer-devel
>