Tangosol Coherence 2.4: The J2EE Availability Myth

Discussions

News: Tangosol Coherence 2.4: The J2EE Availability Myth

  1. Java server applications have gotten a bad rap for availability with sensational headlines like Wily's claim of 88 percent availability. With numbers like these, it's easy to wonder if Java is more capable of "five eights" than "five nines," until you start to consider well-known consumer sites like eBay, Amazon, Orbitz and FedEx that all run large-scale Java infrastructures with extremely good availability numbers.

    To paint a more realistic availability picture using actual production applications, Tangosol has published two customer stories highlighting J2EE applications that provide continuous uninterrupted service running on commodity hardware, achieving massive scalability with no performance degredation.

    IDIMAX provides a web-based service for monitoring and managing remote facilities and intelligent devices throughout the world. They capture real-time status information, proactively initiate corrective actions, provide continuous availability of the associated data with 100% reliability, and analyze historical events and trends. The IDIMAX operating environment is complex, characterized by huge surges as high as 3,000 transactions per second! Even with these requirements, IDIMAX has been able to provide continuous uninterrupted service for its customers, even during the NorthEast Blackout of 2003.

    Dealer.com, another Coherence customer, has experienced 100% uptime without service interruption -- even through an unplanned three-hour database outage that occurred during business hours! Dealer.com is an ASP that hosts 2500 different car dealer sites off of a single Java cluster. Their ASP offerings include a number of back-office transactional and reporting applications for their car dealerships customers. Built entirely in Java, their application runs entirely in-memory in the cluster, providng instant response times. Even their transactional applications have instant response times, because they make extensive use of in-memory transactions with write-behind database caching.

    Asked about the cluster size that it takes to host 2500 dynamic Java web sites, Rob Misek, an account manager for Tangosol, evaded the question by suggesting that "they don’t like us to publicly disclose the exact number of servers that they run this cluster on, but they do state that when they switched to Coherence, their hardware requirements dropped by half, and then over time their requirements dropped by another third because Coherence just kept on improving! Of course, at the same time, they've been growing like crazy, at least doubling the number of sites since they went live with Coherence."

    Yesterday, Tangosol announced its Coherence 2.4 release (see the eWeek article,) which adds support for write-mostly caches (useful for challenges like the temporary archival of massive data feeds with full HA support) and WebSphere 5 HTTP Session Clustering support.

    (A free evaluation copy of Tangosol Coherence can be downloaded from tangosol.com. As always, developer licenses are free, and production pricing is published on the Tangosol web site.)

    Read more: Tangosol Upgrades Data Management Tool for Java Apps

    Threaded Messages (20)

  2. Java server applications have gotten a bad rap for availability with sensational headlines like Wily's claim of 88 percent availability. With numbers like these, it's easy to wonder if Java is more capable of "five eights" than "five nines,"
    Great stories for Tangosol. I wanted to point out though that Willy’s survey of 360 companies across 16 industries looks more adequate sort of claim then 2 hand-picked customer stories. I don’t see how these “collaterals” prove anything beyond two successfully installations of Coherence (which I personally believe is a very decent product)?

    Regards,
    Nikita.
  3. Great stories for Tangosol.
    Thanks, Nikita. I was also hoping that they'd be great stories for the real-world cost-effective high-scale HA (etc.) use of Java and J2EE technologies.
    I wanted to point out though that Willy’s survey of 360 companies across 16 industries looks more adequate sort of claim then 2 hand-picked customer stories.
    For what it's worth, they weren't hand-picked -- they just happen to be the first two stories that got finished. Most of our big customers are not allowed to publish stories; consider that about 70 companies out of the Fortune 100 are already using Coherence, but only a handful are allowed to talk about it. However, we've got several more stories coming; not all have achieved 100% uptime, but they are all running pretty close.
    I don’t see how these “collaterals” prove anything beyond two successfully installations of Coherence (which I personally believe is a very decent product)?
    You're right. This isn't a statistical study at all. On the other hand, these stories show that Java/J2EE can achieve close to 100% availability. With geographical (site) failover, which is officially supported in the upcoming 3.0 release, you can even factor out "act of God," assuming He doesn't get too angry. With careful planning, even "scheduled downtime" can be avoided when upgrading databases and Coherence itself.

    And just to avoid making this just a one-vendor story, there are other HA Java solutions out there that address various types of availability issues, from clustering built into application servers like IBM WebSphere and BEA WebLogic, to messaging systems such as FioranoMQ, Sonic, SpiritSoft and SwiftMQ, to high-availability JDBC solutions for (and some built into) Sybase and Oracle, to the local and global hardware (NOT Java!) load-balancers that make it all possible like F5 and Cisco.

    Peace,

    Cameron Purdy
    Tangosol, Inc.
    Coherence: Clustered JCache for Grid Computing!
  4. consider that about 70 companies out of the Fortune 100 are already using Coherence, but only a handful are allowed to talk about it.
    Wow! It put me wondering of how many of those 360 in survey were also using this great tool ... :-))
  5. infomercial[ Go to top ]

    Personally I log on to TSS to read relavent industry NEWS not be bombarded with "info-mercials" such as this.
  6. infomercial[ Go to top ]

    Same here, This is just marketing dramas!!!. Comparing a cluster enabled, file dependent, hash map circled cache to the top of the software business, do u have anything else buddy??, bring it on. I would like to know how much your cache technology will help me to implement a high volume distributed transactions environment scattered among multiple environment such as J2EE, Mainframe etc.
  7. infomercial[ Go to top ]

    Tangosol, and other JCache solutions, fill a need in the market, otherwise they wouldn't be successful.

    Congrats everyone at Tangosol for this release - looking forward to 3.0!

    cheers,

    Rob
  8. Try it out[ Go to top ]

    Hi t t,
    Comparing a cluster enabled, file dependent, hash map circled cache to the top of the software business
    I have no idea what a "cluster enabled, file dependent, hash map" is. Have you taken a look at Coherence? Our evaluation and development licenses are completely free of charge.

    Coherence is a clustered in-memory data management system that happens to expose its data through the collections APIs, the same APIs that java.util.HashMap happens to use.
    do u have anything else buddy??, bring it on.
    Sure, feel free to take a look at the extensive list of features (on page 2). If that list is not enough, you will have to wait for our 3.0 release ;-).
    I would like to know how much your cache technology will help me to implement a high volume distributed transactions environment scattered among multiple environment such as J2EE, Mainframe etc.
    While Coherence can cluster between mainframe processes and other servers, and manage transactions against its clustered data across all of those systems, it is not (in my opinion) an effective use of the mainframe. Involving an existing mainframe application in a distributed transaction with J2EE systems is typically a difficult and tedious task, and rarely can be accomplished purely with an "off the shelf" approach.

    Later,
    Rob Misek
    Tangosol, Inc.
  9. infomercial[ Go to top ]

    Personally I log on to TSS to read relavent industry NEWS not be bombarded with "info-mercials" such as this.
    you take the words out of my keyboard.

    The projects I've been involved with all had excellent uptime.
    One system ran on 2 machines with no special stuff in between them and was down for less than a day in a year of operation (until the plug was pulled on the project, not for technical reasons but company politics).
    That day had to do with a power failure in the hosting facility...

    I've similar experiences at other projects.
  10. Re: infomercial[ Go to top ]

    Personally I log on to TSS to read relavent industry NEWS not be bombarded with "info-mercials" such as this.
    you take the words out of my keyboard.
    I don't really see the difference between this post and the Spring 1.1 announcement, save that one is a commercial effort touting client successes, and the other in a OSS project.

    I'm sure if these articles were from JCS, we'd just have praise upon praise heaped upon the project.
  11. Re: infomercial[ Go to top ]

    Personally I log on to TSS to read relavent industry NEWS not be bombarded with "info-mercials" such as this.
    you take the words out of my keyboard.
    I don't really see the difference between this post and the Spring 1.1 announcement, save that one is a commercial effort touting client successes, and the other in a OSS project.I'm sure if these articles were from JCS, we'd just have praise upon praise heaped upon the project.
    Me too. I don't see what all the wailing and nashing of teeth is all about. I, for one, am glad to hear when a products works well. I am also pleased to hear that when applications are not available, it has nothing to do with the fact that Java(that includes J2EE and EJBs for those of you who don't know what Java is) was used. So on that note, thank you to all who have posted here saying they haven't had problems whether it was a "simple" web app, or a plethora of servers clustered together.
  12. Re: infomercial[ Go to top ]

    Personally I log on to TSS to read relavent industry NEWS not be bombarded with "info-mercials" such as this.
    you take the words out of my keyboard.
    I don't really see the difference between this post and the Spring 1.1 announcement, save that one is a commercial effort touting client successes, and the other in a OSS project.I'm sure if these articles were from JCS, we'd just have praise upon praise heaped upon the project.
    Me too. I don't see what all the wailing and nashing of teeth is all about. I, for one, am glad to hear when a products works well. I am also pleased to hear that when applications are not available, it has nothing to do with the fact that Java(that includes J2EE and EJBs for those of you who don't know what Java is) was used. So on that note, thank you to all who have posted here saying they haven't had problems whether it was a "simple" web app, or a plethora of servers clustered together.
    It's not the fact that they write about their product, nothing wrong with that.
    But trying to make it look like an unbiassed article when it clearly is not goes beyond that.

    If Microsoft were to write here that their products are good and link to some of their own marketing stories you'd (or many others) be up in arms about it.
  13. Re: infomercial[ Go to top ]

    It's not the fact that they write about their product, nothing wrong with that.But trying to make it look like an unbiassed article when it clearly is not goes beyond that.
    I think anyone who has spent time here knows that Cameron is associated with that product. Yes, would have been nice if his usual signature would have been at the bottom. May not have been his choice. I don't think he tried in anyway to make it look unbiased. For most of what was said, I don't know what being biased has to do with it. Stating facts vs opinion. And there is a link to an eWeek article that is as unbiased as any.
    If Microsoft were to write here that their products are good and link to some of their own marketing stories you'd (or many others) be up in arms about it.
    Maybe. But the difference is Cameron provides a good product. :)
  14. Re: infomercial[ Go to top ]

    Personally I log on to TSS to read relavent industry NEWS not be bombarded with "info-mercials" such as this.
    you take the words out of my keyboard.
    I don't really see the difference between this post and the Spring 1.1 announcement, save that one is a commercial effort touting client successes, and the other in a OSS project.I'm sure if these articles were from JCS, we'd just have praise upon praise heaped upon the project.
    except that the Spring announcement makes no attempt to hide its nature as a product announcement.
    It doesn't pretend to be anything BUT a product announcement, doesn't try to hide behind a veil of being an outside study (for example).
  15. infomercials[ Go to top ]

    Personally I log on to TSS to read relavent industry NEWS not be bombarded with "info-mercials" such as this.
    you take the words out of my keyboard.
    I don't really see the difference between this post and the Spring 1.1 announcement, save that one is a commercial effort touting client successes, and the other in a OSS project.I'm sure if these articles were from JCS, we'd just have praise upon praise heaped upon the project.
    I have no problem with news items intitled "blah releases version n.n of their all-things-to-all-men framework/product" etc etc. What I have a problem with is product release announcements disguised as technical articles or surveys general interest to the community.

    Paul C.
  16. infomercials[ Go to top ]

    I have no problem with news items intitled "blah releases version n.n of their all-things-to-all-men framework/product" etc etc. What I have a problem with is product release announcements disguised as technical articles or surveys general interest to the community.
    We appreciate the constructive criticism and will be sure to clearly state the context in future announcements to avoid any potential perception of disguising the intent of the announcement or misleading the reader. Thank you for your willingness to share your concerns.

    Peace,

    Cameron Purdy
    Tangosol, Inc.
    Coherence: Clustered JCache for Grid Computing!
  17. infomercial[ Go to top ]

    Personally I log on to TSS to read relavent industry NEWS not be bombarded with "info-mercials" such as this.
    Much as I hate to admit it, getting "coverage" on industry sites like TSS is an extremely important part of getting the word out, and that means that there will be a "-mercial" aspect to it, whether one is publicizing an open source project, a commercial software package, or the latest agile consulting craze. The "coverage" is a limited resource, and there are a seemingly umlimited number of entities (companies, projects, authors, pundits ;-) vying for it.

    As for news, we did release a major rev (2.4) of a popular J2EE-related software product, one that (coincidentally) is even used to run this site (theserverside.com) .. so that qualifies as industry news. However exciting the "yet another release" news is, I personally suggested that it would be more interesting (admittedly, what you refer to as the "info-mercial" aspect) to talk about companies achieving extremely high levels of uptime, using Java, particularly accomplishing uninterrupted service despite J2EE and database server failures. Those types of technical challenges keep me up late at night thinking, and make me want to get up in the morning, and there are lots of companies that have those exact requirements, and are looking for reliable and cost-effective solutions.

    Peace,

    Cameron Purdy
    Tangosol, Inc.
    Coherence: Clustered JCache for Grid Computing!
  18. infomercial[ Go to top ]

    Shame on you ¢ameron! How dare you adverti$e your product? Worse yet, you're not even embarrassed about earning money for your efforts. What a disgrace!

    Did you forget that we're the shiny-happy Eloi, the veggie hippies from H.G. Wells' Time Machine? We spend our time basking in the sun and relishing the exotic fruits provided to us for free. We don't taint ourselves with earthly possessions or (gasp!) advertising.

    Oh wait! It looks like the fruits weren't free after all. We're about to be served for a Morlock dinner.

    Sarcasm aside, congrats on your product and keep it up :-)
  19. 88% Availability[ Go to top ]

    It's not the J2EE Application Servers (WebLogic, WebSphere...) responsible for the low availability but the applications developed for these servers.

    IMHO cache products should be employed only after an application is available 99% of the time and the 1% failure is caused by external factors.
  20. How HA is achieved[ Go to top ]

    It's not the J2EE Application Servers (WebLogic, WebSphere...) responsible for the low availability but the applications developed for these servers.
    It's a little bit of everything, like hardware, OS, and JVM failures, but you are right: probably 90% of failures are related to application code. Nonetheless, regardless of the failure type, it's still nice when the end user is not affected (lost sessions, lost transactions, lost data, etc.) when a server crashes.
    IMHO cache products should be employed only after an application is available 99% of the time and the 1% failure is caused by external factors.
    That's an excellent point -- caching itself has little to do with availability. Caching is one of many features that Coherence provides in a cluster:

    1) Coherence manages data in the JVMs that compose the J2EE cluster, and it provides transparent failover for that data without losing any of it. That data can be HTTP sessions, clustered singletons, application data, etc. The result is that the loss of a server doesn't lose any data, so the app doesn't skip a beat, and end users never even know that the server that they were using died.

    2) Coherence provides clustered services. These are not "pinned" services that fail over from server to server, but rather services that are literally spread over the entire cluster in a transparent manner. This eliminates SPOFs (Single Points Of Failure), making the J2EE tier more reliable. These services include caching, data management, lock management and agent invocation.

    3) Coherence can provide transactional data management in the J2EE cluster, assuming that responsibility from tiers behind the J2EE tier, such as the database tier. This allows the application to survive failures in the tiers behind the J2EE tier by queueing the database responsibilies for when the database recovers from a failure. One of the two customer stories we posted witnessed this exact thing with their production application.

    In other words, Coherence is designed first to help applications achieve HA in the J2EE tier, and then optionally (i.e. it may take more work) to provide tolerance for other tiers -- like the database tier -- to avoid their failure from causing the applications that use them from going down.

    Peace,

    Cameron Purdy
    Tangosol, Inc.
    Coherence: Clustered JCache for Grid Computing!
  21. Specific workload[ Go to top ]

    Just to get a better idea of the performance, in case of a workload with 100% data sharing (with lots of updates, obviously), what kind of hit would your latency and thruput get? I am assuming that you ran a test similar to this in-house.

    I am interested in this specific case because Pfister said that DB2 on a cluster used to get 15% overhead (it must be thruput I think), so I would like to compare because that configuration uses RDMA.

    As far as the infomercial criticisms go, I think you are getting it because you are not big enough. When they post some news about Oracle, nobody complains because it's a fact of life. Are you trying to become a fact of life ;-) ?

    Guglielmo