Comments (2)
  • Add a Comment
  • Edit
  • More Actions v
  • Quarantine this Entry

1 enrico.vannini commented Permalink

Hello Joydeep, <br /> more then one month passed since when you started this and .. well.. we have not been very helpful. I feel guilty ... :-( <div>&nbsp;</div> Let me try to give a second startup to this: <div>&nbsp;</div> First requirement: EASINESS. If I have to spend days to get a resource trained to learn something, I will never use it. YOu have better number then I have, please tell me: is this what has happened to Tivoli Performance Analyzer ? May be ... <br /> In the real life you will never care of complex algorithm, because they will never match the requirement #1 above. I recently visited a customer that to get a baseline out of past response time, was taking the avg resp time at that time in the last 7 days (i.e. now it is 3.08, Wednesday, let's get avg 3.05 to 3.10 since last Wed to today), discarding highest and lowest, and averaging them again. No matter if the formula is not mathematically correct, and if a statistic guru would cry listening to it. Their system was working fine, even because on the long term, a not correct formula will tend to be good anyway. <br /> We discussed about the baseline algorithm used by ITCAM AD on TEP, with distribution buckets for response time. But the gap in easiness was so huge .... <br /> I think ITM is following this principle with its baselining style, and may be ITCAM TT is doing the same too, even easier because it is automatic, no human intervention. Not the case for ITCAM for WAS. <br /> But yet from the above line you can see the second point we should avoid: here above I mention 3 different solution I could use to baseline WAS response time. The 3 solution are all from the SAPM Tivoli portfolio and they conflict to each other ... <br /> Even if I spoke of baselining, i.e. to get a threshold out of past values, same concepts of course applies to analytic, to the capability to predict what will happen in the next week if we keep this trend. <div>&nbsp;</div> hope this help getting this discussion stareted <div>&nbsp;</div> Regards, <div>&nbsp;</div> Enrico <br />

2 JoydeepBanerjee commented Permalink

Hi Enrico - <div>&nbsp;</div> It is never too late to get a discussion started! You make some very good points. <div>&nbsp;</div> Ease of usage is of primary importance. In some domains where advanced analytics are greatly used like HealthCare, Finance there are statistical/mathematical gurus sitting there doing the number crunching but in our domain this is not prevalent at all. Our mission is to make it as simple and consumable as possible. <div>&nbsp;</div> The other issue is related to baselines. While some of our tools do allow the user to create baselines, I think your view that it be done automatically is absolutely the way the industry trend is going. Probably there will still be time and place for users to create their own baselines using tools on single variables like response time but it is not fair to place this burden on the user anymore. Modern systems are far too complex. <div>&nbsp;</div> Now let’s talk about IBM products that you refer to. IBM Tivoli Performance Analyzer was one of our early initiatives into the world of predicting trending. It predicted how a variable will look like in 10 days from now based on the past history. The dynamic baseline capability of IBM Tivoli Monitoring (ITM) on the other hand enables monitoring to be done using dynamic thresholds (the dynamic thresholds being set based on past behavior). You make a very good point Enrico when you talk about different approaches to baseline by ITM and IBM Tivoli Composite Application Management for WebSphere (ITCAM for WAS). While totally agreeing with you that the concept of response time buckets etc introduce in ITCAM for WAS could be simplified and made more consumable (a lesson learnt), one of the main reasons why ITM baselines could not used here is a little more complex though. Due to very high volume of the transactional data generated from WebSphere, the baseline in ITCAM for WAS happens prior to the data entering ITM environment. ITM therefore gets only aggregated data for WebSphere response time. Now you could legitimately argue that we could build baselines from the aggregated data too. Sure you could (and baseline from aggregated data is done all the time), but I have not compared the results from the two to comment on if one is better than the other. <div>&nbsp;</div> Again Enrico, thanks for taking the time to respond. And thanks to all of you who have read the blog but understandably did not have time to write down from your thoughts! <div>&nbsp;</div> Thanks <br /> Joydeep