Welcome!

Silverlight Authors: Automic Blog, Michael Kopp, AppDynamics Blog, Kaazing Blog, Steven Mandel

Related Topics: Weblogic

Weblogic: Article

Diagnosing Tough Performance Problems

Diagnosing Tough Performance Problems

Although many of the symptoms of performance problems (e.g., poor response time) are similar throughout the application life cycle, the underlying causes and the techniques used to diagnose them become more complex in later stages as the load increases and the configuration becomes more complex. In this article, we discuss the tools and techniques that are useful in diagnosing tough performance problems that occur under realistic high loads. We also illustrate why tools used in development or under limited load conditions are not suitable for finding such tough performance problems.

Reproducing Performance Problems Under Load
There are two principal kinds of performance problems:
persistent problems that affect the performance at all times, and transient problems that occur intermittently and for a limited amount of time. The former are typically easy to reproduce as they can be triggered by high enough workload. The latter are harder to reproduce because the right configuration, state, and workload are required to reproduce them. To diagnose tough performance problems under high load, we need a reliable way of reproducing the problem and the ability to examine the internals of the application under load.

Persistent performance problems (e.g., the Empty-Cart transaction always has poor response time) are often reproduced using load-testing tools. In contrast, current techniques for reproducing transient problems (e.g., response time of the entire system quadrupled for 15 minutes at 2:00 p.m. yesterday) involve guesswork and ad hoc testing in order to approximate the original configuration and load in the production environment. Using current approaches, it may take anywhere from weeks to months to reproduce and diagnose transient problems. An alternative is to use real workload technology, which records the production workload (including all transaction requests and responses) when the transient problem occurs, and plays it back in the test environment in order to reproduce the problem. This reduces the time to reproduce and ultimately solve these problems from weeks and months to hours and days.

Diagnosing the Root Cause
Once the performance problem is reproduced, diagnosing the root cause requires drill-down analysis that correlates external symptoms with potential root causes. This involves:

  • Finding the bottleneck or the transactions, beans, servlets, and methods that consume the most time
  • Breaking down the aggregate time spent in a type of transaction across the servlets, beans, and methods it uses
  • Correlating individual transactions from an HTTP request, through servlet and bean calls, and down to JDBC calls and SQL statements to find the chains of invocations that produce slow response time
  • Examining individual thread execution profiles to see which threads were the bottleneck and where the time was spent

    Profiling tools used by developers often provide aggregate information (e.g., statistical summaries of calls to methods) that helps in such analysis. However, they don't isolate individual calls or objects, nor do they gather information about arguments and results. Additionally, profiling tools that are based on the JVM profiling interface (JVMPI) suffer from a significant overhead (between 200 and 1,000%). This performance impact makes loading the application nearly impossible and significantly skews results. Profiling tools based on JVM sampling have lower overhead (around 100 to 200%) but provide even less information. Such tools can be used under higher load but are less helpful in diagnosing tough problems. Byte code instrumentation technology can provide an arbitrary level of detail, and its overhead can be managed by limiting the scope of instrumentation without sacrificing the level of detail. When properly implemented, gathering data using byte code instrumentation can have overhead as low as 5 to 50%.

    Production monitoring tools instrument a small subset of Java methods and report summary information on beans and methods that take the most time.

    These tools discard much of the detailed information required for root-cause diagnosis because they sample method calls and aggregate results in order to reduce overhead. While such monitoring highlights application bottlenecks in production, it does not go the proverbial "last mile." Developers still invest significant time and effort to reproduce the bottleneck in the test lab to diagnose and fix the underlying problem.

    Detailed diagnostic tools are used in test environments to capture the details of each significant method call. Such tools fall into two categories: tools that aggregate and summarize the captured information online for presentation, and tools that save detailed information to disk for offline analysis. While both types of tools help find bottlenecks and provide a breakdown of aggregate transaction time, the second category also supports correlation of individual transactions through different software layers and examination of the execution profiles of individual threads. This analysis of individual transactions and threads is critical to diagnosing complex performance problems.

    In summary, diagnosing tough performance problems in J2EE applications requires a reliable way of reproducing the problem and a low-overhead technique to examine the internal details of the application under load. The capture and playback of real workload from production provides the best option for reproducing difficult performance problems. Performance diagnostic tools based on byte code instrumentation can capture and correlate individual method calls and thread invocations with low overhead and provide the best option for drill-down analysis.

  • More Stories By Ashutosh Tiwary

    Ashutosh Tiwary has 12 years of software development and performance consulting experience at Boeing, Hewlett-Packard, and Teknekron Communications Systems. He is a PhD
    candidate in computer science at the University of Washington, where his dissertation work forms the basis for Performant's technology.

    More Stories By Przemyslaw Pardyak

    Przemyslaw Pardyak codeveloped Performant's core technology and has eight years of research and development experience including performance management. He is a PhD
    candidate in computer science at the University of Washington.
    pp

    Comments (0)

    Share your thoughts on this story.

    Add your comment
    You must be signed in to add a comment. Sign-in | Register

    In accordance with our Comment Policy, we encourage comments that are on topic, relevant and to-the-point. We will remove comments that include profanity, personal attacks, racial slurs, threats of violence, or other inappropriate material that violates our Terms and Conditions, and will block users who make repeated violations. We ask all readers to expect diversity of opinion and to treat one another with dignity and respect.


    IoT & Smart Cities Stories
    With the introduction of IoT and Smart Living in every aspect of our lives, one question has become relevant: What are the security implications? To answer this, first we have to look and explore the security models of the technologies that IoT is founded upon. In his session at @ThingsExpo, Nevi Kaja, a Research Engineer at Ford Motor Company, discussed some of the security challenges of the IoT infrastructure and related how these aspects impact Smart Living. The material was delivered interac...
    Intel is an American multinational corporation and technology company headquartered in Santa Clara, California, in the Silicon Valley. It is the world's second largest and second highest valued semiconductor chip maker based on revenue after being overtaken by Samsung, and is the inventor of the x86 series of microprocessors, the processors found in most personal computers (PCs). Intel supplies processors for computer system manufacturers such as Apple, Lenovo, HP, and Dell. Intel also manufactu...
    Darktrace is the world's leading AI company for cyber security. Created by mathematicians from the University of Cambridge, Darktrace's Enterprise Immune System is the first non-consumer application of machine learning to work at scale, across all network types, from physical, virtualized, and cloud, through to IoT and industrial control systems. Installed as a self-configuring cyber defense platform, Darktrace continuously learns what is ‘normal' for all devices and users, updating its understa...
    At CloudEXPO Silicon Valley, June 24-26, 2019, Digital Transformation (DX) is a major focus with expanded DevOpsSUMMIT and FinTechEXPO programs within the DXWorldEXPO agenda. Successful transformation requires a laser focus on being data-driven and on using all the tools available that enable transformation if they plan to survive over the long term. A total of 88% of Fortune 500 companies from a generation ago are now out of business. Only 12% still survive. Similar percentages are found throug...
    OpsRamp is an enterprise IT operation platform provided by US-based OpsRamp, Inc. It provides SaaS services through support for increasingly complex cloud and hybrid computing environments from system operation to service management. The OpsRamp platform is a SaaS-based, multi-tenant solution that enables enterprise IT organizations and cloud service providers like JBS the flexibility and control they need to manage and monitor today's hybrid, multi-cloud infrastructure, applications, and wor...
    The Master of Science in Artificial Intelligence (MSAI) provides a comprehensive framework of theory and practice in the emerging field of AI. The program delivers the foundational knowledge needed to explore both key contextual areas and complex technical applications of AI systems. Curriculum incorporates elements of data science, robotics, and machine learning-enabling you to pursue a holistic and interdisciplinary course of study while preparing for a position in AI research, operations, ...
    CloudEXPO has been the M&A capital for Cloud companies for more than a decade with memorable acquisition news stories which came out of CloudEXPO expo floor. DevOpsSUMMIT New York faculty member Greg Bledsoe shared his views on IBM's Red Hat acquisition live from NASDAQ floor. Acquisition news was announced during CloudEXPO New York which took place November 12-13, 2019 in New York City.
    Codete accelerates their clients growth through technological expertise and experience. Codite team works with organizations to meet the challenges that digitalization presents. Their clients include digital start-ups as well as established enterprises in the IT industry. To stay competitive in a highly innovative IT industry, strong R&D departments and bold spin-off initiatives is a must. Codete Data Science and Software Architects teams help corporate clients to stay up to date with the mod...
    Tapping into blockchain revolution early enough translates into a substantial business competitiveness advantage. Codete comprehensively develops custom, blockchain-based business solutions, founded on the most advanced cryptographic innovations, and striking a balance point between complexity of the technologies used in quickly-changing stack building, business impact, and cost-effectiveness. Codete researches and provides business consultancy in the field of single most thrilling innovative te...
    Atmosera delivers modern cloud services that maximize the advantages of cloud-based infrastructures. Offering private, hybrid, and public cloud solutions, Atmosera works closely with customers to engineer, deploy, and operate cloud architectures with advanced services that deliver strategic business outcomes. Atmosera's expertise simplifies the process of cloud transformation and our 20+ years of experience managing complex IT environments provides our customers with the confidence and trust tha...