Performance tuning DynaTrace

Hi Steven

I assume these 5 clients are opening dashboards that include things like PurePaths, Database Statements, Exceptions, ... - so - data from PurePaths? If that is the case then please make sure that these dashboards are mainly used for specific diagnostic use cases, e.g: finding out why in a certain timeframe transactions were slow. It doesnt make sense to e.g: "Show me all PurePaths of the last 24 hours". This is root cause for causing really high CPU but also Memory Usage on the dynaTrace server because the server needs to load all these PurePaths from disk into Memory.

It should be no problem at all to have clients looking at charts because this data comes from the Performance Warehouse. Depending on the number of measures you have in the warehouse and the sizing of your database it could still be a problem. Therefore please have a look at the Health Dashboards that are available through the Start Center. Here is the link to the doc: Deployment Dashboards

In the upcoming dynaTrace 6.0 release we address the problem of many clients trying to do PurePath-based analysis. We are currently doing an Early Access Program. In case you have the time and a system where you can install dT 6 let us know and we can sign you up for the early access program

Andi

Hi Steven

I saw the attachment you uploaded. It explains the problem you are experiencing. You are opening a Business Transaction Dashlet and trying to get the result of the last 7 days. In that case the dynaTrace Server needs to load all PurePaths of the last 7 days to produce that table. The reason being is becauset the Busienses Transaction Dashlet then allows you to drill to the individual PurePaths. That is GREAT for doing targeted diagnostics - BUT - only makes sense if you can first narrow down the timeframe to not have the dT Server load milions of PPs

I assume that you would like to see a chart that displayes e.g: Response Times of each individual Message Type over the last couple of days, weeks or months. To do that simply create a Chart and add the Response Time measure of your Business Transaction on the Chart. That will work SUPER FAST as the data comes from the Performance Warehouse and is not computed on demand by looking at all PurePaths.

If you then look at the chart and find hotspots you can zoom into that timeframe and from there drill to the PurePaths for further diagnostics

I hope this makes sense

Thx Andreas,

so it seems by adding the purepath duration measure (also purepath cpu duration) dynaTrace goes for this to the disk instead off the database?

First explaining why we used purepath duration?

This is an example of web service calls, we see that the response time is 0.X ms instead of X seconds, we think that this is because of

async call, in the purepath details we see that the execution logic time is into the purepath duration and not into the response time, so we added the pure pathduration measure, (for the messages dahsboard, purepath duration = response time so i willl directly change this), but for all of our async web call we have a problem because we cannot use repsonse time, Correct? Or are there other solutions to get the total web call duration out of the database instead of using the disk,

Thx

Thx,

So everything which is into a time chart graph is retrieved from the datawarehouse, at the moment we ask data from a 'table dashboard' we retrieve a lot pure path data from the disk.

Ofcourse i don't know the technical challenges of system,, but i taught a table dashboard would also be saved into the database and in a later time be aggregated. 

We have a company culture that everyone want to see all the data  

Of course that is not possible, but i need to make a compromise,

So thinking on a solution:

-->  Now i can only place heavy dashboards on separate account, only used by a few people

      Example my boss wants to compare database statements from today with statements of yesterday.

      So i need to say to him that this is a action that will cost time, correct? Because we saw times of 10 mintutes to retrieve query stats of 6H

      We just get data from disk. Correct?

     It would be interesting that i give table dashboards to people BUT that they cannot select more then 30 minutes each time.

     This would certainly be accepted, Is there a way to achive this? Or is this a Feature Enh

--> i can control my dashboards (rights), but each chart has a drill down so it is possible that everyone uses a chart to drill down on exceptions in a 

     time period of 6H data, possible to prevent?

I can ask people to select only 30 minutes of detail data, but they will get exceptions, query's, ... on 2 week data, 

i used to be a developper / tester and believe me they will do this,

"let see how long the system last"  or just out of a unawareness point of view ,

so i have to prevent this somehow 

Is this correct? If we cannot prevent huge data selection, the dynaTrace server will go to 100% cpu (or has memory issues),

which blocks the data insert, which gives use buffer overflow and data loss? Or is the write / read path separated / secured?

How do other companies do this?

Thx

Just a max time selection coupled to a dashboard would make a huge difference

The memory problem on the dynaTrace server is addressed in the upcoming dT 6 release. I am hosting a Whats New Webinar in 2 weeks where we show these new capabilities.

If you like to see things in tables I have three things for you

a) Dynamic Measure Matrix Dashlet -> check it out. It allows you to show multiple measures in a table

b) Business Transaction Hotspot DAshlet. This also displays Business Transaction Data per splitting but instead of going through all PurePaths it is taking the data from the Perf Warehouse -> should have told you that right from the start 

c) The Chart Dashlet also has a table on the bottom of the chart -> not sure if that is what your users like - but - well - it is a table

Andi

Hello,

I have tried the Business Transaction Hotspot DAshlet, it is a huge difference, 72H in 5 seconds.

I aim happy to see that i still have a link to the purepaths , how is that possible?, 

What is the advantage of the Original dashboard compared to the  Hotspot DAshlet?

If a select the same period on both dashoards (today) then

i see a small difference in stats (example count & duration)?. At this moment i will not worry about this and

 quickly change all the dashlets, made into the poc, to the Business Transaction Hotspot DAshlet

One more question? Only related to pure^paths, Is it somehow posssible to force the user to select a small time period,

for example no more then 30 minutes, if the user want more data, just 'scroll'' 30 minutes back /  further,

any workaround, or is this just a FeatureE

Many thx for the solution