JProfiler 6.0 released


News: JProfiler 6.0 released

  1. JProfiler 6.0 released (10 messages)

    JProfiler 6.0 has been released by ej-technologies. Among the highlights in JProfiler 6.0 are reworked thread profiling, a new method statistics view, exceptional method run analysis, tree maps, auto-tuning for instrumentation and support for FreeBSD x86 and Linux PPC. A comprehensive overview of the new features is available on the What's New page A fully functional trial download for 10 days is available.

    Threaded Messages (10)

  2. Great tool[ Go to top ]

    I have been using JProfiler for quite some time and it is a great tool for profiling (imho the only one in its league) So I'm very interested to see what this release is going to bring. Peter Veentjer
  3. Re: Great tool[ Go to top ]

    Well Peter then JXInsight must be in a completely different time dimension than all others. [in relation to performance profiling and not heap memory dump analysis] Seriously its best to ignore what Peter and I just said and instead evaluate each profiler against a largely code base with known issues (performance cases) - some obvious (low hanging fruit) some less so obvious (multi-resource bottlenecks, low high performance micro-tuning). Of the 4 tools I routinely test 1 of them will rarely identify the second cases because its tracing overhead is gigantic, 2 will identify some of them and 1 will identify nearly all of them and that one runs in production as well. Amazingly at least to those only familiar with one or two products most give completely different hotspot sets for the second cases. The overriding concern in choosing a tool should be the quality, accuracy and coverage of the data collected and that is largely based on the runtime overhead and how both the software & system execution behaviors are mapped to a performance model. If you are just using a profiling tool in an ad hoc manner to find low hanging fruit then the above is largely irrelevant - any tool even primitive ones shipping with the JDK can help. But if you need to squeeze as much as possible out of a well tuned code base and direct tuning work based on production data then you seriously need to understand what and how a product instruments, measures and collects its data. William
  4. Re: Great tool[ Go to top ]

    By the way if you need to quickly verify someones claims of being a "performance expert" simply ask him to list out the typical costs ranges of various instrumentation & collection techniques on a per method call/execution basis per metric/meter/measurement basis - contextual and non-contextual. if the response given is in percentages then clearly he or she is a fool or charlatan. Most of the so called Java performance experts I have talked with would fail this test but yet they keep running around promoting tools without this knowledge. Luckily for them most of the issues they run into fall into the "low hang fruit" category and most importantly during development and early testing. The rest of the issues well they last for months sometimes years in production.
  5. Os license jxinsight?[ Go to top ]

    Hi William, You wont hear me saying that i'm a performance expert, but i have been using an open source license of jprofiler on multiverse, a software transactional memory implementation, on almost a daily basis. And have been able to reduce lot of performance problems because with the right settings, there is very low overhead, so you are not hunting ghosts. This year i also followed a dynatrace performance tuning course, and such a product is much more usefull if you have an 'enterprisy' environment and need to get closer to the production environment. Imho jxinsight belongs more in the same ballpark as dynatrace. But if you are willing to donate an open source license to the multiverse project, i could be placing positive comments on your tool as well. Peter Veentjer
  6. Re: Os license jxinsight?[ Go to top ]

    Actually we go beyond this by instead offering up our performance expertise (time) following a submission of performance test kit to us. No need to learn about our unique to performance engineering based on activity costing, multi-resource metering and strategy chaining. We have recently done this for JBoss RESTEasy. We send you back a summary of the key hotspots in your system without us even having to look at your code though we usually take a look at the code to verify our findings and propose ways this can be improved but at the end of the day performance insight comes when you have both the data and the model. The model is largely in your head we provide the data and the bridge between the model you assume and the actual model itself. data + model = insight
  7. Re: Os license jxinsight?[ Go to top ]

    Peter I am a little bit confused by your usage of JProfiler in the context of your open source project. Surely one cannot effectively tune a STM solution using a time based (execution or sampling) profiling solution. The cost of performing a single clock counter read is at least 70 nanoseconds on a standard unix system. Make two of these and then update a performance model and you are close to half a microsecond from then on you are effectively tuning at the microsecond level. For such systems I would expect the performance engineering focus should be on using path analysis along with count based meters (field read/write counts, call counts,...). This is much more effective during analysis of the software execution model. This is how we tune our own probes runtime. We offer the only solution on the market that does this at whatever level of execution abstraction - below code profilers and above high level (low coverage) path tracers/journalling solutions.
  8. Cheap profiling[ Go to top ]

    Hi William, i use jprofiler in combination with some load generating tests to see where time is spend in the internals of the stm It doesnt help me to write more efficiënt stm datastructures but it helps me to locate overhead caused by the stm implementation. We are working on an event based profiler to get transaction level and atomicobject level statistics (which transactions fails often with a writeconflict or which atomic object caused it for example). But this is still in à poc fase because there is way too much overhead. So if you have any tips for doing very cheap event based profiling
  9. Re: Cheap profiling[ Go to top ]

    Hi Peter, Is this load test (code, scripts,...) also made available with the sources? Have you published performance reports using jprofiler so your findings to date? If you send me links and the setup can be done pretty quick I will get performance results prepared for you under various configurations of JXInsight Probes using various hotspot metering strategy settings, meters, probe providers and instrumentation sets. We do this type of work because we are not just a product company. We specialize in offering performance "war council" services to some of the largest software companies in the world which explains why our product is extremely extensible and packed full of features devoted to tracking down the hardest (in terms of detection) of performance problems in production systems. We do not sit in some room think up product features or copying features from other products we are out in the field pushing this technology to the limits and extending in ways no one could really dream off unless presented with a challenge. Already having inside knowledge of various products and open source projects helps in such engagements though it does mean you are less appreciative of popular products used in the Java world. Ignorance can be bliss - sometimes.
  10. Re: JProfiler 6.0 released[ Go to top ]

    Congrats! As peter already said, this is the best tool for some serious profiling! I am curious if there is now a more detailed of the memory. Especially memory inspections like the ones which are in yourkit (see ) This would really help me a lot with classifying the memory usage (e.g. string pool usage, many objects, size of the objects etc)
  11. Re: JProfiler 6.0 released[ Go to top ]

    Can it track call trees across JVMs/tiers (e.g., servlet calling EJBs doing JPA/JDBC to a remote database etc.)?