SLA on Reports

How do you determine an SLA (service level agreement) for reports performance? Lets say a report returns 10,000 rows with about 50 columns in it. What is an acceptable time that we should target? Any ideas?

thanks
Arvind.

Ask the business what the business requirement is. Software and hardware can be purchased to meet most reasonable SLA's these days. When I run a report I don't like waiting around more than a couple minutes. I'm rarely using the BI tool and accessing the database directly though. For most canned reports, I would expect 10 seconds or less to return data.

Thank you for the response. When you say canned reports, are you saying a report that allows prompts and then pulls the data based on those prompts?

If the business defines exactly what they are looking for, then it is much easier for the IT to deliver. But the company that I work for, the business's only business is to whine. They don't give us proper requirements in the first place, let alone helps us setting up an SLA. This is the reason why I am curious to find out if there is an acceptable response time when it comes to canned reports, analytical reports etc.

Thanks again for your response.

You don't work for the same company as the business? Walk in their shoes for a day. How long would you want to wait?

I should have known. When I call others whiners, that will make me a whiner.

My question was what is an acceptable SLA or what is the standard used in the industry, if there is one. I think you answered it as 10 seconds for canned reports, which is good. Lets say if the canned report allows the user to select a region and we establish an SLA of 10 seconds for running the report for one region. Now the business wants to select 15 regions and expect it to be done in 10 seconds.

As you mentioned since I work for the same company, should I try to walk in their shoes by recommending the business to invest in technology etc to make it happen in 10 seconds or should I negotiate the SLA with the business that their expectation is unreasonable.

I've worked on both the business and IT side of this equation and I agree that a reasonable response time is within 10-15 seconds.

In my opinion, the first step is to determine what the business is looking for. Is it a canned report, a canned report with prompts, multiple prompts with multiple selections?

Once you have determined the requirements, you should be able to estimate the run time within a QA environment.

It's not reasonable to provide an SLA regarding response time if you do not fully understand what the business is looking for. It's common for the business requirements phase to be the most difficult part of the process.

Any form of SLA comes down to a cost/benefit excersise. So, the business may say, 'We need to have this report done in x (minutes, seconds, whatever...). The IT response should not be to challenge it, but rather accept it and respond (after some analysis), what would be necessary to achive the desired service level. So, it becomes, "Yes we can do it, but here is what it will cost to do so, and why...". You may also provide less costly alternatives along the lines of "if y is an acceptable time frame, we can achieve that with the existing infrastructure" and, if y is way off, some in-between choices at different costs. It then becomes a business decision... is it worth it?

Thank you for your insight into this. I appreciate it.

Just to give a little background, in the company I am working for there is no concept of SLA. They dont have one for their regular applications that generates revenue let alone the the BI apps. There is a perception that "everything is slow", but despite repeated attempts, I am unable to ascertain specific pain points from the business.

Given this situation, I want to do something proactively from the BI/DW Architecture perspective. It is one thing if the Business sets the goal and asks the IT executes it. But my situation is different I guess.

Regarding the canned reports, 10-15 seconds is acceptable for a report that prompts and choose a single selection and extract it. I can use that as a basis and then build upon that. Again this one region could be a couple of thousand rows or it could be over 40-50k rows of data. The response time would vary drastically in these two cases. What I am looking for is, "if you are downloading an X MB of data from the DW via BI interface, you can expect the report returns in y seconds".

Is this something practical? Once again, I want to set a benchmark and test my reports response against it. I am just lost what that benchmark is.

Arvind.