There remains an art of good cross charging.  It involves providing the services to the business at a price they are willing to pay, whilst making sure your costs are covered (the art of cross charging).  What remains difficult with grid though is the chargeback.

Let’s say I’ve got an application on grid, it needs 350 blades, with windows, grid licenses, oracle client, power everything. The business then want to add another application to the grid.

A polite IT department will charge you on utilization, and quite possible find it very difficult to cover their costs without large amounts of debate. Say the cost of the grid is £500,000 start up and £200,000 a year.  Application A uses the grid 20% of the time, so it pays 20% of the cost,  Application B is going to use the grid 15% of the time, so they pay 15% of the cost. Very courageous, very good.

But I can respond but my application is restricted to 70 blades, so don’t I pay 20% of the 15%, that’s 3% right?. The maths gets complicated,  how do I charge the depreciation?  Doesn’t that make my anual cost with depreciation £367,000 a year?

The other way to do it is to keep life simple. The business pay per blade, regardless of whether it’s in use, they want the grid functionality with 350 engines available, they pay for 350 blades, it’s up to the applications and application code being able to use the grid efficiently.

It’s not the most efficent way of doing things, but until all applications remain equal, the blade requirements in terms of operating system, layered components are equal (or compatible), this remains an effective way.  It’s say £500,000 installation costs, license costs etc, that’s depreciated over say 3/4 years (whatever your standard is), say it’s 3 years. 

That’s about £167,000 a year depreciation, plus your set fee for supporting the server infrastructure, (finger in the air time) £250 a box per year (£87,500), plus possibly a grid team cost (though you might just hide that cost internally). Or round figures £254,000 per year grid cost billed to the business line.

In essence in this model, you want more capacity for your new application, well either use your existing grid, or buy more blades, the utilization and how applications pay for it, becomes an application related issue. This does cause grid and sever growth to increase, however, until the applications and the grid become compatible with each other, managing the cost per cpu per hour gets complicated unless you’re using a compute on demand service where you actually know the hosting costs, the power etc.

It remains a challenge which the IT departments debate on an ongoing basis….

Tweet This Post 

This entry was posted on Thursday, January 11th, 2007 at 9:05 am and is filed under Blades, Grid Computing. You can follow any responses to this entry through the RSS 2.0 feed. You can leave a response, or trackback from your own site.