-1

Response time increases largely from 100 to 150 users in load test.also we see constant 75 % utilization of processor in 150 while 100 it reaches 75 % twice during steady state comes down to 65 %.the large increase in response time in 150 user test only seen for update transaction.respone time increases 3 to 4 times (30 sec to 100 sec fpr update txn).read transaction have minor increase 3 to 4 seconds in 150 user load test.how to proceed with analysis to find root cause

Performance testing 2011
  • 1
  • What you are describing sounds like you are hitting a limit - possible also contention / locking - are you saying that a single update takes 30-100 seconds? or the whole workload takes that long? What is the database, and whats it running on. That will dictate the tools available. Oh, and make sure your indexes are set up correctly as that may fix any problems. – TomC Oct 15 '19 at 03:27
  • Database is sql server and running on windows 2012 server.we see many full scans and latch wait per second is high in Perfmon counters. Do we need to analyze traces.full scan are reaching more than 15 to 20 per seconds.we can also deadlocks.also lock request per second is between 60 to 80. – Performance testing 2011 Oct 15 '19 at 09:07
  • Of everything you said there, the deadlocks sounds the most worrying. I would have a look at the application code and try and identify where they are coming from. Everything else can be fixed, by tuning, configuring, or upgrading. – TomC Oct 15 '19 at 21:27

1 Answers1

0

The reasons could be in:

  1. Your application is overloaded in terms of CPU, RAM, Network, Swap or Disk usage. Consider monitoring above (and application specific metrics if any) using an APM tool or 3rd-party solution like JMeter PerfMon Plugin
  2. Your application or its middleware components (application server, database, load balancer, etc.) are not properly configured for high loads. Refer to the individual components documentation and follow performance tuning tips to ensure that they are configured to utilize all available resources
  3. It might be the case that your application code itself is the bottleneck, consider re-running your test having profiler tool enabled and inspect the results looking for the slowest functions, heaviest DB queries, largest objects, etc.
Dmitri T
  • 159,985
  • 5
  • 83
  • 133
  • Database is sql server and running on windows 2012 server.we see many full scans and latch wait per second is high in Perfmon counters. Do we need to analyze traces.full scan are reaching more than 15 to 20 per seconds.we can also deadlocks.also lock request per second is between 60 to 80.app server mwmory and cpu looks normal.the increase in response time is only for save,submit and add – Performance testing 2011 Oct 15 '19 at 09:11