Sun produces J2EE Application Server High-Availability paper

Discussions

News: Sun produces J2EE Application Server High-Availability paper

  1. Sun has released a paper on J2EE Application Server High-Availability. The paper discusses the quest of supporting an application availbility of 99.999% (The five-nines). The analysis discusses various configurations of system architecture, using application servers, and HADB nodes (reliable db architecture for session replication).

    The paper happens to use Sun ONE Application Server 7, but is relevant for all servers.

    It is interesting to learn that of the various configurations analyzed, 4 instances of the application server and 4 instances of the HADB for replication gave the highest availability number. This is compared to configurations with 1, 2, 6, 8, and 10 instances.

    It is an interesting read if you care about highly available enterprise applications. The analysis used the RAScad availability analysis tool. RAScad is a Web-based RAS architecture modeling tool for system design and development phases.

    Read: Availabity Modeling Technical White Paper

    Executive Summary

    Reliability, availability, and serviceability (RAS) are becoming increasingly important for networked computer server and storage systems running business-cretical applications. RAS is one of the major issues considered in designing and deploying enterprise products and services. This paper presents an availability analysis for multitier, Java and Web-based application deployments running on the Sun ONE Application Server 7, Enterprise Edition software. The results show that this software is able to support 99.9999% application availability in common web deployment scenarios.

    Threaded Messages (17)

  2. Flame: Avaliability[ Go to top ]

    They should not talk:
    http://biz.yahoo.com/rf/031017/tech_sunmicro_s_p_4.html
    above says:
    "Moody's cut Suns debt rating to 'Baa3' from 'Baa1,' one notch above junk status".
    Would availability mean ECC memory or ... that company is stable?
    Bye, bye SlowLaris.

    For a company with no market share, they do a lot of press releases. Here is the market share of iPlanet/SunOne/lets call it something else so people forget how bad it is and think it something else. How big is the Sun Java Sofware Market share? :
    http://news.netcraft.com/archives/2003/04/10/java_servlet_engines.html
    That is how much Java market will shrink by when the last Sun person leaves.

    I use J:Rockit VM, others use IBM VM. Sun VM has StringBuffer memory leaks and you are not allowed to dstirbuted the VM. J:Rockit lets me include the VM with my product.

    Lesson for other companies: You should listen to users, like IBM and MS do.
    EJB+BluePrints+JSF, right.
    Avalability... are they for real.

    .V
  3. Flame: Avaliability[ Go to top ]

    Vic: I use J:Rockit VM, others use IBM VM. Sun VM has StringBuffer memory leaks ..

    The Sun VM does not have StringBuffer memory leaks. You (or libraries that you use) may misuse the StringBuffer class and hold onto instances of it, which means that whatever memory it is using will not be garbage-collected, but don't blame that on Sun. That's a problem with the programmer using the StringBuffer.

    Vic: .. and you are not allowed to dstirbuted the VM. J:Rockit lets me include the VM with my product.

    The JRE is redistributable the last time I checked.

    Vic: Lesson for other companies: You should listen to users, like IBM and MS do.

    Now that is something that I can agree with ;-)

    Peace,

    Cameron Purdy
    Tangosol, Inc.
    Coherence: Clustered JCache for Grid Computing!
  4. I work with Sun ONE App Server 7[ Go to top ]

    Hello everyone,

    I have worked with all three versions of Sun ONE Application Server 7. It's a really good product. I am not going to bash other app servers, but this one normally rate #3 or #4 on a number of polls. The Platform Edition is free for development and production. The Enterprise Edition that this paper talks about is only $10K/CPU, which is prices pretty good. I know a number of organizations using it include Dept. of Defense.

    Thanks
    Mike Brown
  5. I work with Sun ONE App Server 7[ Go to top ]

    I agree Sun ONE App Server 7 is very good. Even the "Free" version is better than some Commercial (i.e. you have to pay for) app servers (e.g. JRun4). IMHO. SunOne Studio5 (the default IDE... has become very stable and fast and has good integration with SunOne Appserver 7 so development & deployment is quite easy as well).

    Cheers,

    Smythe
  6. Cameron,

    According to Heinz Kabutz (javaspecialists.co.za) there was at least a different semantics in JDK 1.4.1 StringBuffers that could lead to memoryleaks in code that was previously safe.

    The linked newsletter contains a reference to a bug in the JDK bugparade.

    Joost
  7. memoryleak in JDK using StringBuffers[ Go to top ]

    The memory leak problems in StringBuffer have been solved in the past minor releases.

    Check http://developer.java.sun.com/developer/bugParade/bugs/{4224987,4637640,4724129,4295713}.html and bugs marked as "related". They are all fixed.
  8. Flame: Avaliability[ Go to top ]

    They should not talk:

    > http://biz.yahoo.com/rf/031017/tech_sunmicro_s_p_4.html
    > above says:
    > "Moody's cut Suns debt rating to 'Baa3' from 'Baa1,' one notch above junk status".
    > Would availability mean ECC memory or ... that company is stable?
    > Bye, bye SlowLaris.
    >
    > For a company with no market share, they do a lot of press releases. Here is the market share of iPlanet/SunOne/lets call it something else so people forget how bad it is and think it something else. How big is the Sun Java Sofware Market share? :
    > http://news.netcraft.com/archives/2003/04/10/java_servlet_engines.html
    > That is how much Java market will shrink by when the last Sun person leaves.
    >

    No,read the report. It specifically states that the numbers are incomplete; excluding (partly or wholly) sites running IIS, NES, SunONE or Apache makes the report almost useless. Judging Sun's (Java) software business by the the size of the JavaWebServer market (which was EOL'd Feb 2001) is not that smart - using it to back up an argument on a public forum - priceless
  9. Re: Flame: Avaliability[ Go to top ]

    I would like to add one more thing to this Web Server issue. Sun's Web Server is not free. Apache has great numbers because for every cnn.com there are 1,000 mom-and-pop.coms. Netcraft's data also shows the Sun's Web Server is #1 within the Fortun 100. Pock around on Netcraft and you'll find a lot big name companies use this sever. I think that it still shows up on Netcraft has "Netscape Enterprise", but it's the same product. Everyone from cnn.com to washingtonpost.com to playboy.com use it.

    Thanks
    Mike
  10. This is a really informative read...[ Go to top ]

    Someone has reaally done their homework - especially with the state models.

    I learned a great deal, and hope to see a similar benchmark come out of the JBoss camp (my preferred Application server right now because of 1) the price and 2) the fact that the same version can be installed on all platforms.

    Thanks to those who put the time into a fantastic white paper.

    Best,

    John C. Dale
  11. Don't call it 5 nines[ Go to top ]

    From page 12

    “Several aspects of a physical deployment are beyond the scope of the current model. To simplify the study, the following elements are assumed not to fail:
    • Database and directory servers (Oracle, LDAP, and so on)
    • Web server and load balancer
    • Electrical power
    • Network communication
    • HVAC (Heating, ventilation, and air conditioning)

    We assume that in a typical data center well-engineered and commercial-quality solutions for the above elements are already in place.”

    These aspects reduce real life availability of their model to 3 nines.

    I'm looking for good information on how to plan for better than 3 nines availability. Does anyone have recommendations?

    Gary
  12. Re: Don't call it 5 nines[ Go to top ]

    Gary,

    I fail to understand why you don't think that this is 5x9's. All of those items listed have solutions for reliability. Do you expect your application server to come with a battery backup?

    Many software application claim 5x9's and they don't cover power failure, etc. Why do you expect this product to handle these situations?

    Mike
  13. Re: Don't call it 5 nines[ Go to top ]

    How can you promise less than 5 minutes of downtime per year if you can't recover during a 12 hour blackout? Even if other vendors make similar claims it will mislead people into believing a single site server cluster will have 99.999 percent uptime.

    Availability during infrastructure breakdowns is expensive. It would really be helpful to see how cost effective availability tactics are that keep a service up while infrastructure is down.

    Gary
  14. Re: Don't call it 5 nines[ Go to top ]

    In the interest of full disclosure I do work for Sun and did have little part to play in the production of the paper in question.

    My guess is that anyone architecting a service that requires continuous availability would realise that the total availability is the product of the availability of all of the elements in the system - power, building, storage, network, etc. One of the intentions of the paper is to help those people understand what the capabilities and limitations of the app server tier is so they can build a realistic availability model for the entire system they are building. We're not trying to convince people this is pixie dust (TM) that will make their systems immune to tsunami, blackouts and earthquakes.

    Any vendor claiming support for "High Availability" without actually providing a robust model explaining how it is achievable isn't really helping people build such systems.

    Caveat Emptor.

    - Rich


    > Availability during infrastructure breakdowns is expensive. It would really be helpful to see how cost effective availability tactics are that keep a service up while infrastructure is down.
    >
    > Gary
  15. 5x9s holds[ Go to top ]

    Gary,

    How do you recover from a 12 hour blackout? You have a generator. You can have 5 minutes of downtime per year if you build your architecture according for their specs. It doesn't mean that everything needed is included with the app server. The servers are not provided, but you'll need them.

    ETrade and Putnum use Sun's application server (not 7, yet). They have little or no down time. How do they do it? You build your architecture is replication in different geographic locations, or use battery backup for short blackout and then have generators kick in. It's good that Sun provides a checklist of other items to be consider.

    It's still 5-9s of reliability
    Mike Brown
  16. No EJB's[ Go to top ]

    Remarkably, the applications used in this analysis do not use EJB's. I find this very interesting.

    //ras
  17. No EJB's[ Go to top ]

    Remarkably, the applications used in this analysis do not use EJB's. I find this very interesting.

    >
    > //ras

    7.0 EE does not support failover for SFSBs (though it does recover handles to EJBs stored in HttpSession); 7.1 will add support for, among other things, SFSB failover, RMI/IIOP load balancing / failure detection as well as MQ broker clustering.

    - Rich
  18. No EJB's[ Go to top ]

    Rich,

    Can you contact me via email at cpurdy at tangosol dot com .. I'd like a chance to talk to some of the study authors. Thanks!

    Peace,

    Cameron Purdy
    Tangosol, Inc.
    Coherence: Clustered JCache for Grid Computing!