Slow startup (JIT) times for ASP.NET applications
I have several asp.net applications that I want to monitor with DynaTrace. Without DynaTrace these applications startup in 10s of seconds. By startup I'm referring to the time it takes for the just-in-time compiler to compile about 30 dlls and 100 aspx, asmx, etc. pages. With an agent deployed, with what I would call an optimized approach the time grows to several minutes. We have recorded times in excess of 4 minutes.
How can I determine which sensor(s) are causing the most delay? The only course of action I can think of is start placing them one at a time, which, could take a while.
Hi Robert
4 minutes sounds a bit too long. Where is your Collector deployed? The collector is doing the actual instrumentation and should therefore be close to the agent in order to avoid slow startup because of network latency.
If you have the Collector close to your agents I suggest sending in your logfiles through a Support Ticket.
Andi
hi Robert
based on what you describe, i would guess it's related to either
- network latency: as described by Andi above
- NGEN: do you use NGEN heavily for your application?
- for correct .NET application diagnosis, we have to disable NGEN which may lead to (usually slightly) longer startup times
- excessive instrumentation: please have a look at Agents Overview -> Deployed Sensors (on the bottom) to get an idea of the number of instrumented method
if disabling (unplacing, _not_ setting inactive) all sensors helps decreasing your startup time, it will most probably be related to #1 or #3.
best, Christian
Hi, I was wondering a little bit about this as well.
How much overhead does dynatrace generally introduce in JITing of methods / classes and so forth?
Hi
During startup our .NET Agent needs to send the .NET Assemblies that get loaded by the CLR to the dynaTrace Collector which is doing the actual byte code instrumentation. The "overhead" or slowdown in that startup phase is therefore depending on two factors
a) how many assemblies get loaded into your application
b) the network connection between your agent and the collector
A best practices for deploying dynaTrace is to put Agent and Collector as close as possible to avoid this type of slowdown. But - you will definitely experience some slowdown. This is however ONLY during startup but not during regular operation
Andi
Is there any way possible to check how much overhead this generates?
We have a customer with an archictecture that makes the application do JIT A LOT.
So getting infromation about this would be appreciated.
There is some information in the agent log files. It tells you how many classes were sent to the collector to be instrumented and how long that took
Do you have an example?
