Home

News: GigaSpaces 6 announced, allows POJOs in JavaSpaces via Spring

  1. GigaSpaces 6 has been released, rebranded as XAP (eXtreme Application Platform). This release is notable for its OpenSpaces technology, which leverages Spring to provide POJO support to the JavaSpaces API. In addition, XAP supports JMS, JDBC, and JCache as standard APIs. The release structure of GS6 is similar to the prior releases, with a community edition, a clustering edition, and an enterprise edition. The OpenSpaces API allows the deployment of a "processing unit," which allows clustered usage of POJOs without any invasive coding in the POJOs themselves. A processing unit is deployed through the XAP server and systems that participate in the grid are immediately able to manage processes and data at need defined as service levels in the XAP server. Therefore, if a grid is configured to send processing across the grid if CPU loads climb above 75% (or RAM usage climbs above a certain level, etc.), any participant on the grid will start handling processes or data as the system determines availability. Congrats to GigaSpaces on the GS6 release - TSS has been working with it and it's a very nice implementation of the space-based programming model. GigaSpaces XAP 6.0 will be generally available in September 2007.

    Threaded Messages (42)

  2. how this can be compared to Oracle coherence data grid? is it a data grid?can i run data application on the cached data when the db is down? what is the main deference between XAP and the JavaSpaces in the blitz and RI(opensource)?why should i pay u money? what are its pros against terracotta jvm level clustering
  3. how this can be compared to Oracle coherence data grid?
    is it a data grid?can i run data application on the cached data when the db is down?
    what is the main deference between XAP and the JavaSpaces in the blitz and RI(opensource)?why should i pay u money?
    what are its pros against terracotta jvm level clustering
    GigaSpaces is closer to a compute-grid than a data grid, Tangosol is closer to a data grid. They both overlap enough to create some confusion to most people not experienced in this area. You can argue either as being a solution to many grid problems and the choice is often a religious one. Can you run with the database down? Well it depends on your SLAs, you can configure it either way, the answer is yes if you want it to. Difference between XAP and "classic" JavaSpaces? XAP IMOHO is closer to Spring than classic JavaSpaces. When you think that Jini is older than J2EE and that J2EE has had several spec revisions it's only right to accept Jini's JavaSpaces to progress in its own way. A few years ago Blitz was good news for GigaSpaces because without it GigaSpaces was almost a proprietary API. There was Intamission (now defunct) and Sun's rather crap and extremely complex reference implementation but Blitz came into the lead in the open source market thanks to funding by Paremus (still a very interesting company) they two of them (GigaSpaces and Blitz) became very viable JavaSpaces implementations. The future was however unclear, hooking in to the J2EE API seemed rather daft as J2EE was already on the downward slope in 2004 so we came up with Spring as a future API. GigaSpaces still retains its pure JavaSpaces API but it's far more flexible use through Spring. Terracotta's JVM level clustering is very limited in functionality when compared to GigaSpaces however there are problems that it solves quite well and it's well worth a look into. I hope that helps, -John-
  4. GigaSpaces is closer to a compute-grid than a data grid, Tangosol is closer to a data grid. They both overlap enough to create some confusion to most people not experienced in this area. You can argue either as being a solution to many grid problems and the choice is often a religious one.
    As Max the miracle man said, this is mostly true. Here at GigaSpaces, we are focused on three grid aspects: Data, Events and Compute. GigaSpaces provides full and complete support for all three. What we have found along the way is that solving only one or two of the problems is not enough to build a highly performant solution. As an example, wouldn't it be nice to have your domain model form the basis of your event driven architecture (instead of using JMS) and then act on them (ala Data Grid). What we have done with version 6.0 is to enable simplified development model based on Spring allowing Spring developers to feel right at home when developing a highly transactional system. Event driven is similar to Spring 2.0 JMS support and data grid access is similar to Spring ORM wrappers. Another benefit of GigaSpaces 6.0 is allowing to take this Spring application and deploy it onto GigaSpaces SLA driven containers. We provide a lightweight Grid deployment of Spring applications based on certain SLAs (such as relocate a certain deployment if the memory passes a certain threshold, or breaks a user defined monitor). What we again noted (as many other Spring developers) that a full blown JEE container is not required when using Spring, especially when combined with GigaSpaces (note that I am not talking about web containers here). The above two paragraphs shows that GigaSpaces actually focus on solving a broad problem of building a highly performant application. From the get go of developing your solution within your IDE using simple constructs and programming model (Spring + GigaSpaces Spring components) to deploying it to a large scale production system. Personally, I am of the opinion that such a solution forms the basic requirements for any "next gen" application servers/container. With more and more "AJAXified" web applications and real time data requirements, building any other solution just feels too much like J2EE before Spring. This is what we have tried to do with Open Spaces (and there is much more to be done, stay tuned...). Cheers, Shay Banon Compass Founder System Architect at GigaSpaces
  5. Terracotta's JVM level clustering is very limited in functionality when compared to GigaSpaces however there are problems that it solves quite well and it's well worth a look into.

    I hope that helps,

    -John-
    Doesn't seem proper to bring up Terracotta in this context (but I will do it as well :P ). Specifically: 1. Gigaspaces is Jini / Spaces and Terracotta is POJO WITH OBJECT IDENTITY INTACT. As you pointed out, Terracotta can solve problems like HTTP session clustering w/o serialization, distributing a simple Java collection used as a cache w/o rewriting code to add serialization or putting updated values back in the collection, not to mention Java thread-based coordination such as java.util.concurrent that Gigaspaces cannot do. 2. Gigaspaces does "the grid thing" with proprietary APIs whereas Terracotta does Master / Worker with java.util.concurrent in order to solve a similar challenge. (BTW, if you want to judge the definition of POJO, taking a look at Owen Taylor's excellent blog, referenced elsewhere in this thread will take you to a Gigaspaces-based Master / Worker to which you can compare Terracotta.) 3. Terracotta's plumbing differs completely from Gigaspaces leaving their scalability characteristics to differ drastically. Example: Javaspaces in general can easily be extended to do distributed query where Terracotta's fundamental approach is "network attached memory" (NAM) leaving distribution to the developer. Another example: Terracotta's NAM makes HTTP session apps scale linearly since it sends only field-level deltas to the cluster where Gigaspaces will scale roughly on the order of Weblogic session clustering. In other words, they are good at different things. As you point out John, the 2 solve different problems. But it is not really worth confusing everyone by even mentioning them together. The 2 are going after different goals. Terracotta is meant to provide high availability for Java applications in the enterprise, be they 2 nodes or more with lots of OSS such as EHCache, Rife or Wicket, Tomcat or Jetty, Lucene, etc. or a simple use of java.util.concurrent to divide workload amongst a grid. Gigaspaces--well--XAP tells me by its name that it is "extreme" meaning proprietary APIs and approaches (ignoring their Spring façade for a moment) which are well suited to solving a workload distribution problem. Excellent stuff, I am sure, but not really an "either or" decision with Terracotta. Thanks, --Ari
  6. As you point out John, the 2 solve different problems. But it is not really worth confusing everyone by even mentioning them together.
    --Ari
    Ari, It wasn't me who bought up the two in the same paragraph it was the chap I was answering
    what are its pros against terracotta jvm level clustering
    Obviously you're the man with the answers on Terracotta and I look forward to getting a deep-dive in Barcelona if you're there. Best regards, -John-
  7. As you point out John, the 2 solve different problems. But it is not really worth confusing everyone by even mentioning them together.
    --Ari

    Ari,
    It wasn't me who bought up the two in the same paragraph it was the chap I was answering
    No worries. I just wanted to make it clear that we have invested over 40 man-years in DSO over the past 10 months and things are constantly evolving. There is no way you could keep up with all we have going on, unless you are using the product regularly. I hope to convince you to do just that, over time...

    what are its pros against terracotta jvm level clustering

    Obviously you're the man with the answers on Terracotta and I look forward to getting a deep-dive in Barcelona if you're there.

    Best regards,

    -John-
    I welcome the opp. to catch up real soon. I will try and make it out to TSS Barcelona. Was great fun, last year!
  8. As you point out John, the 2 solve different problems. But it is not really worth confusing everyone by even mentioning them together
    Judging by the type of applications discussed at our London user conference two days ago such as reconciliation application presented by Smart-Stream, real-time analytics presented by BOFA, pure .net global (WAN) trading application by trader-tools, Order Management system by Virgin and BAT and also data intensive analytics by Commerce-Bank it is fair to say that we are addressing different problems. We're looking at providing a complete solution for scaling-out of stateful application and provide data-grid as a component of our proposed architecture as Shay Banon outlined in his earlier post. It was apparent that for those users the main value in using GigaSpaces was with the ability to achieve true linear scalability of their *entire* application since we address all the application bottlenecks: both the data and more importantly the architecture. Non of those users could even consider using a caching solution such as terracotta to achieve that goal (I'll get back to that later on). What we found throughout the years is that the majority of the cases having a data bottleneck problem is a symptom of scalability issue. What lead most people to look into caching solution in the first place is actually scaling requirement. Since In most stateful applications the database is the most obvious bottleneck it makes sense to put a caching solution in-front of it to reduce the overhead and therefore raise the bar on how much transaction one can push through the same system. That path only improve the throughput of the application but doesn't really address the scaling of the entire application i.e. it doesn't get you to the point where you can build you application in a way that scaling will be achieved simply by adding more instances dynamically based on SLA without changing your configuration or application code. With the new GigaSpaces XAP release we made it so simple such that one can be as Scalable as Google and keep the application simple as Spring. Having said that we realized right at the beginning that not everyone can look at scaling their entire application at the get to go. For that purpose we provide the ability to use our DataGrid as a separate edition of our product at a significantly reduced price. We also provides a free community edition for those that are looking for L1, L2 cache or as we refer to Master/Local cache. The comparison with other caching solution such as Ehcache or Terracotta is relevant only under the context of our DataGrid edition and not XAP. Under that context I would say that at the very high level I would use the commodities OSS caching tools such as Ehcache and even tTerracota for simple scenarios such as http Session sharing and mostly read caching scenario of relatively small datasets and gain performance through replication. Under this context it would be nice if a solution like Ehache or Terracota could be open to plug-in other implementation and act as a common façade for those different implementation. For real transactional application that requires heavy read/write requirements, SQL query support and need to scale in both capacity (couple of Terra bytes of data) and throughput using smart partitioning - those type of solution wouldn't be a good fit and that probably explains why we haven't really came across any of them in real competitive situation. When they did came out they where often ruled out at the early stages of the evaluation. There were plenty inaccurate statements to say the least made by Ari in his post with regards to proprietary approach as well as some other comments on other features that I don't want to even bother responding at this stage. My recommendation when this type of communication happens has always been don't rely on anyone biased information (I'm including myself on this category) simply take a look at the documentation and try out the two approaches and products. I do have some reservation on the conceptual differences between the two approaches that the two products represent i.e. byte code manipulation, server centric cluster etc. I'll probably address them separately on individual post. See you all in Barcelona. Nati S. Write Once Scale Anywhere
  9. The comparison with other caching solution such as Ehcache or Terracotta is relevant only under the context of our DataGrid edition and not XAP.
    Nati, First and foremost, I agree that people can just download and play so let's not get stuck here trying to convince folks via rhetoric. I do need to clean up your misconception where you called Terracotta a "cache." I could see why you are confused. Terracotta clusters EHCache for folks who are already using that framework. We spend a lot of time with Greg Luck over at EHCache and he has been pointing us in a fruitful direction regarding EHCache + Terracotta integration. That might look like Terracotta is similar to or part of EHCache to someone not actually using either EHCache or Terracotta in a production system. No worries. --Ari
  10. Hi Joe, Data Grid is a term that encompasses many of the functions of this solution but GigaSpaces defines further characteristics in terms of the architectural and programming model offered in this release --Rather than me repeating everything here, please take a look at this: http://www.gigaspaces.com/wiki/display/GS6/Welcome+to+GigaSpaces I believe with all respectable Data Grids and similar solutions, you can elect to run application logic against the system when the DB is down, this is part of the nature of the reliable and transactional aspect of these solutions. GigaSpaces6.0 absolutely supports this model if you choose it. GigaSpaces XAP is mainly different from many other things in that it is an implementation of a TPC-architecture which stresses Transparent Partitioning and Colocation of work, events and information. Additionally, the Spring-based Processing Unit concept and support for a broader set of APIs plus the use of POJOs, PONOs, and POCOs really sets this product apart. You should pay GigaSpaces money : ) so that GigaSpaces can continue to research and produce production-quality middleware products that come complete with award-winning documentation, training, and support. - not enough? OK: Additionally, the GigaSpaces solution is unusual in its clustering support and range of functionality - check out the Query capability offered against the Objects held in memory. (see: this link for more specifics) Again, there is a wealth of value and functionality available in this product - it is best to refer to the docs on the GigaSpaces website or perhaps take this offline if you like. . . Owen Taylor Senior Director - Worldwide Technical Communications GigaSpaces Technologies Web: www.gigaspaces.com Blog: www.gigaspacesblog.com My Blog: http://jroller.com/page/owentaylor owen@gigaspaces.com
  11. I think GigaSpaces should change its policy with regard to the comunity edition. Space-based computing has not caught up much momentum or has meanwhile lost the momentm it once had. To attract more interest it would be usefull to have a community edition developers could really play with. The community edition limited to a single node here is a problem in my opinion. For me there would be no problem to have a community edition with only a limited number of features or limited to only a small number od nodes. But if the thing is from the beginning limited to a single node it is only good for evaluating the product at best. And this does not do much about revive space-based computing. Regards, Oliver
  12. I think GigaSpaces should change its policy with regard to the comunity edition. Space-based computing has not caught up much momentum or has meanwhile lost the momentm it once had. To attract more interest it would be usefull to have a community edition developers could really play with.

    The community edition limited to a single node here is a problem in my opinion. For me there would be no problem to have a community edition with only a limited number of features or limited to only a small number od nodes. But if the thing is from the beginning limited to a single node it is only good for evaluating the product at best. And this does not do much about revive space-based computing.

    Regards, Oliver
    Hi Oliver Thanks for the comment and interest. To be clear. With the community edition you could run cluster of applications using the master/local topology which essentially provides you the ability to have a cluster of local caches that are syncronized through central space. The central space can be made persistent for high availability. Will that suit your needs for the small scale deployment? Nati S
  13. Hi Nati, I don't understand some of the terms you are using. So I can't really evaluate what you are saying. I would be happy enough if in the Community Edition you could have a couple of client Java VMs that access the GigaSpace server from other machines than from the one the GigaSpace server is installed on. I don't know whether that is the case since it's not clear whether a client machine accessing the GigaSpaces server on a different machine counts as node or not. Regads, Oliver
  14. Oracle Coherence Data Grid[ Go to top ]

    how this can be compared to Oracle Coherence data grid?
    I don't believe that the two can be compared directly. When it comes to Data Grids, Coherence is unparalleled. Performance-wise, we helped a customer port an application from Gigaspaces and the result was around a 200x improvement in transaction performance (over 200 times the transactions per second on the same hardware, and transaction latency was significantly better). That said, performance is a distant third behind the two primary areas of focus that Coherence has: Reliability and Scalability. It's important to understand that reliability is an absolute prerequisite for scaling stateful applications; in other words, if you can't count on data being both continuously available AND correct, you cannot scale a stateful application. Without attempting to put down any other products, I will simply point out that Coherence is very well known for its exceptional reliability. As far as actual usage, a recent analyst report (sponsored by one of Gigaspaces' investors) put the pre-Oracle market share for Coherence at about 8x that of Gigaspaces, and (considering we're a couple years younger in the market) that gap is only widening.
    is it a data grid?
    See: http://www.jroller.com/page/cpurdy?entry=wtf_is_a_data_grid
    what are its pros against terracotta jvm level clustering
    Simply put: Terracotta isn't clustering. Terracotta is a client/server architecture with a central server. That's as much a cluster as any client/server SQL database server circa 1995. While I am not a fan of Gigaspaces' approach to clustering, since it is not an organic cluster (i.e. it cannot provide deterministic behavior for stateful applications during and after failure conditions), it is clustering nonetheless. On the other hand, if Terracotta solves your specific problems, then does it really matter if it is clustering or not? Just make sure that you are comfortable with the not-exactly-open-source license and that you don't need to buy support, since that support will apparently cost you more than just buying the alternatives. Peace, Cameron Purdy Oracle Coherence: The Java Data Grid
  15. Re: Oracle Coherence Data Grid[ Go to top ]

    Cameron -- Is this the Cameron we know, or have you been hijacked by Corporate aliens? If it truly is you, it looks like you've adopted some of your new company’s tactics: bashing others rather than focusing on your technology and product. I would have expected such behavior from a sales guy (as we‘ve already witnessed from other sales guys in the field), but not from someone who is a technology thought-leader. I hope it’s not too late to bring back the old Cameron we all love and that you haven’t completely moved to the dark side… Throughout the life of GigaSpaces, we've been in situations in which competitors have chosen to throw dirt at us as part of their sales strategy when they felt cornered. The guiding principle to our team is to never respond in kind, focus on what we’re good at, and let the customer decide for himself. I don’t intend to change that principle now. The best proof for our success is the hundreds of deployments and the glowing public testimonials we’re getting from our customers. The latest is from the Gallup Organization: "GigaSpaces has saved us significant time and cost. We have already identified four major advantages since deploying GigaSpaces: the elimination of capped memory constraints; improved reliability; a platform with more options for innovation than ever before; and a very flexible messaging and job processing infrastructure." "The GigaSpaces platform handles all of these issues. In fact, the benefits of GigaSpaces' approach have been so widely felt within The Gallup Organization that the company recently awarded GigaSpaces a Gallup Partner Excellence Award. Gallup bestows the honor on Gallup partners who provide excellence in service to Gallup associates and clients." I will also point you to our London User’s Conference that ended last week in which 6 different customers presented their use cases and gave testimonials about their results and benchmarks. Interestingly enough, quite a few of them evaluated alternative products… expect these named customers and their presentations to be available on our web site in a few days. What have you done with the old Cameron ? we want him back… If anyone is interested in more specific information/responses regarding these and other claims being made by any of our competitors, please feel free to contact me directly at: natis AT gigasapces.com. Nati S. Write Once Scale Anywhere
  16. Re: Oracle Coherence Data Grid[ Go to top ]

    .. bashing others rather than focusing on your technology and product.
    No bashing intended. I was just highlighting our focus on the Data Grid technology, and one result of that focus. Since the original application that I mentioned was architected by Gigaspaces (using the "space based architecture") and running on the latest version, I felt it was a fair comparison, i.e. what one can achieve with a Data Grid instead of a space based architecture, which in this case was more than a 200x speed-up. To be fair, there are many functional areas that Gigaspaces focuses on that we don't do, e.g. service grid stuff that competes with Platform and DataSynapse, and since we do not try to do those things, I would not try to compare to those.
    I hope it’s not too late to bring back the old Cameron we all love ...
    Come on Nati, I have to occasionally say _something_ that is at least a _little_ controversial ;-) Besides, I am sure we'll be working more closely together over time, as we work to standardize some of this grid stuff and drive the market forward as a whole. I still don't see any reason why these products can't be working together. Peace, Cameron Purdy Oracle Coherence: The Java Data Grid
  17. Re: Oracle Coherence Data Grid[ Go to top ]

    Hi Cameron That sounds more like the Cameron i used to know :)
    I was just highlighting our focus on the Data Grid technology, and one result of that focus.
    Without knowing the exact details behind your claims I'm sure that any reasonable person would understand that such figures doesn't make real sense. Many of our existing customers had witnessed the exact opposite results, well to be fair not the exactly the opposite but we where slightly better in terms of throughput in most scenarios. You can find one of the public comments on that regard here For those interested in the truth and feel confused our product ships with an out of the box open-source benchmark that can be easily plugged with other implementation and can give you the real results in a matter of minutes. The same goes with market share argument, our figures for last year shows the opposite. So to sum it up i would say that from a pure DataGrid perspective our product has proven to be first of its class with unique features such as support for SQL Query, true Continues (SQL) Query, JPA like POJO annotation and mapping, enahanced integration with external data source, as well as SLA driven DataGrid and Management and monitoring tools. (Keeping the honesty) The only issue that we heard with previous release was complexity which has been significantly addressed through our Spring integration and enahanced documentation for which we even been recognized for Innovative Wiki Documentation. Note that since were looking at a broader context our aim was to simplify the entire application architecture and not just our DataGrid configuration therfore 6.0 is quantum leap on that regard since it makes the entire application development and deployment as simple as Spring! An important aspect of it is testability. Our aim was to enable users to develop their application on a single node for functionality tests and move to production without changing their code or even configuration.This enables simple iterative development which results in shorter development cycle, better quality and productivity etc.
    there are many functional areas that Gigaspaces focuses on that we don't do, e.g. service grid stuff that competes with Platform and DataSynapse, and since we do not try to do those things, I would not try to compare to those.
    Let me clarify - we do not see our selfs in competition with DataSynapse or Platform in fact we have mutual accounts as well as partnership agreements and we are working on few mutual accounts these days as well. Were more focused on scaling out stateful low latency applications which a very different domain. Position wise our eXtreme Application Platform (XAP) provides to those applications the equivalent of what application-servers provides to web applications. Those interested in more details about our DataGrid and how it fits into existing Enterprise Grid environment i would suggest reading the following whitepaper : Data-Awareness and Low-Latency on the Enterprise Grid
    Besides, I am sure we'll be working more closely together over time, as we work to standardize some of this grid stuff and drive the market forward as a whole. I still don't see any reason why these products can't be working together.
    I agree - i look forward for it Will you be on TSS Barcelona event or Spring ONE this week? Nati S.
  18. Price list?[ Go to top ]

    Just make sure that you are comfortable with the not-exactly-open-source license and that you don't need to buy support, since that support will apparently cost you more than just buying the alternatives.
    Is there a price list for Coherence?
  19. Re: Price list?[ Go to top ]

    Is there a price list for Coherence?
    Yes. It is on the Oracle price list. Peace, Cameron Purdy Oracle Coherence: The Java Data Grid
  20. Re: Price list?[ Go to top ]

    Is there a price list for Coherence?


    Yes. It is on the Oracle price list.

    Peace,

    Cameron Purdy
    Oracle Coherence: The Java Data Grid
    I'm not seeing it on here: http://www.oracle.com/corporate/pricing/pricelists.html While you're all reading this thread, do any of you have licensing or pricing plans in place or in the works for an on-demand deployment model, like Amazon EC2?
  21. Re: Price list?[ Go to top ]

    ...

    While you're all reading this thread, do any of you have licensing or pricing plans in place or in the works for an on-demand deployment model, like Amazon EC2?
    You can use Terracotta for free on EC2. I became a total fan of EC2 a while ago: http://orionl.blogspot.com/2007/02/terracotta-cluster-on-amazons-elastic.html The best thing about it is that, with the EC2 API, you can spin up new Terracotta nodes as needed programatically. I've been meaning to set up a few EC2 machine images with Terracotta pre-installed so people can try it out for themselves. This seems like the right time to do it. I'll post back when I have something to try.
  22. EC2[ Go to top ]

    ... do any of you have licensing or pricing plans in place or in the works for an on-demand deployment model, like Amazon EC2? ...You can use Terracotta for free on EC2...
    EC2 does sound interesting, but, I don't believe they are ready yet. Today, you can get a fourth of a P4 class machine for about $.10/hr, but, nothing if you need large dedicated capacity. For instance, our "Grid" customers want to test with atleast 1000 concurrently running members. I guess, you could do basic validation to see if a certain kind of software works with a certain node count, but, you cannot do any type of benchmark testing to prove "performance" software. I have been told larger machines will be available in EC2 in November or later. Eagerly awaited. Meanwhile, we, at GemStone, have significantly increased our investments to create our own Infiniband enabled lab with hundreds of nodes. Cheers! Jags Ramnarayan http://www.gemstone.com http://jagslog.blogspot.com/
  23. Re: EC2[ Go to top ]

    ... do any of you have licensing or pricing plans in place or in the works for an on-demand deployment model, like Amazon EC2? ...You can use Terracotta for free on EC2...


    EC2 does sound interesting, but, I don't believe they are ready yet. Today, you can get a fourth of a P4 class machine for about $.10/hr, but, nothing if you need large dedicated capacity. For instance, our "Grid" customers want to test with atleast 1000 concurrently running members.
    I guess, you could do basic validation to see if a certain kind of software works with a certain node count, but, you cannot do any type of benchmark testing to prove "performance" software. I have been told larger machines will be available in EC2 in November or later. Eagerly awaited.
    Meanwhile, we, at GemStone, have significantly increased our investments to create our own Infiniband enabled lab with hundreds of nodes.

    Cheers!
    Jags Ramnarayan
    http://www.gemstone.com
    http://jagslog.blogspot.com/
    I'm not sure what your point is... a grid is all about using commodity hardware to scale out horizontally, right? EC2 is giving you virtual commodity hardware and, even better, letting you provision as many as you need when you need them. This seems like a perfect match for grid vendors. I'm a customer starting up a business who doesn't want to spend lots of money on hardware, hosting, networks, etc. I want to be able to scale my infrastructure on demand and eventually in response to changing usage patterns. So sell me... I think Amazon should open their pricing and billing as a service and let 3rd party vendors tie in with your usage data to do their own billing. That way I could pay for what I use to Amazon and to the vendors whose software I'm licensing.
  24. Re: EC2[ Go to top ]

    Hi Jason
    ...grid is all about using commodity hardware to scale out horizontally, right? EC2 is giving you virtual commodity hardware and, even better, letting you provision as many as you need when you need them. This seems like a perfect match for grid vendors.
    It's interesting that you bring that up. We actaully saw that comming and already got instances of GigaSpaces running on EC2 platform. I believe that a classic target for this type of platform will be Web 2.0 applications that need the scalability to deliver AJAX, RSS feeds etc. The economic model for those applications fits nicely into the $/cpu. Another interesting model that fit nicely with it is the SASS (Software As A Service) i.e. instead of downloading GigaSpaces, installing it tuning it you can get access to pre-installed XAP or XAP-DataGrid and just use it. This is where our SLA driven container fit-in nicely - it provides facillity for managing multi-tenancy environment in which different application instances can be running over a shared environment and access different instances of DataGrid without interrupting with each other. As you can imagine this type of management capabilities are critical enabler for running in any utility based computing environment. The following paper provides good background on how the SLA driven DataGrid works in that type of environment: Data-Awareness and Low-Latency on the Enterprise Grid Feel free to contact us if your interested in further details. Nati S. GigaSpaces - Write Once Scale Anywhere
  25. Re: EC2[ Go to top ]

    I'm not sure what your point is... a grid is all about using commodity hardware to scale out horizontally, right?
    My point is simply that as a vendor in the Grid space, some of our customers are looking for very large scaled tests, literally a thousand nodes (4 CPU nodes) dedicated to a GemFire data fabric test. This will come close to their production environment in the fuutre. My understanding is that EC2 is not ready for such tests.
    ...I'm a customer starting up a business who doesn't want to spend lots of money on hardware, hosting, networks, etc. I want to be able to scale my infrastructure on demand and eventually in response to changing usage patterns. So sell me...
    We might be talking two different things. Your interest is to primarily host your application that can scale horizontally using EC2. Might very well fit your needs. Our motivations are different: How to enable a customer who wants to validate the vendor technology quickly? EC2 will work for certain kinds of tests, but, not all. That is why we designed a decent Grid lab to help customers carry out their tests for pilot projects on our dime. Cheers! Jags Ramnarayan http://www.gemstone.com http://jagslog.blogspot.com/
  26. Re: EC2[ Go to top ]

    I'm not sure what your point is... a grid is all about using commodity hardware to scale out horizontally, right?


    My point is simply that as a vendor in the Grid space, some of our customers are looking for very large scaled tests, literally a thousand nodes (4 CPU nodes) dedicated to a GemFire data fabric test. This will come close to their production environment in the fuutre. My understanding is that EC2 is not ready for such tests.

    ...I'm a customer starting up a business who doesn't want to spend lots of money on hardware, hosting, networks, etc. I want to be able to scale my infrastructure on demand and eventually in response to changing usage patterns. So sell me...

    We might be talking two different things. Your interest is to primarily host your application that can scale horizontally using EC2. Might very well fit your needs.
    Our motivations are different: How to enable a customer who wants to validate the vendor technology quickly?
    EC2 will work for certain kinds of tests, but, not all.
    That is why we designed a decent Grid lab to help customers carry out their tests for pilot projects on our dime.

    Cheers!
    Jags Ramnarayan
    http://www.gemstone.com
    http://jagslog.blogspot.com/
    And those are your only potential customers? What about small companies that want to start out right and be able to scale linearly as they need to? Seems like it would be short-sighted to ignore them. SOMEONE is going to get this stuff set up and let people play with it. It's like being a VC: You give it to 10 startups and if 1 or 2 make it and grow to be big, you've got a great reference account and a solid revenue stream. What's your cost if the startup folds? Zero.
  27. Re: EC2[ Go to top ]

    What about small companies that want to start out right and be able to scale linearly as they need to? Seems like it would be short-sighted to ignore them.
    No debate here. Projects that make do with the EC2 capabilities should find it quite useful. I believe we have created (or in the process) a EC2 AMI with simple scale out tests.
  28. Re: Price list?[ Go to top ]

    I'm not seeing it on here: http://www.oracle.com/corporate/pricing/pricelists.html
    +1. This is just naive curiosity. Are there an prices yet?
  29. Re: Oracle Coherence Data Grid[ Go to top ]

    Simply put: Terracotta isn't clustering. Terracotta is a client/server architecture with a central server. That's as much a cluster as any client/server SQL database server circa 1995.

    While I am not a fan of Gigaspaces' approach to clustering, since it is not an organic cluster (i.e. it cannot provide deterministic behavior for stateful applications during and after failure conditions), it is clustering nonetheless.

    On the other hand, if Terracotta solves your specific problems, then does it really matter if it is clustering or not? Just make sure that you are comfortable with the not-exactly-open-source license and that you don't need to buy support, since that support will apparently cost you more than just buying the alternatives.

    Peace,

    Cameron Purdy
    Oracle Coherence: The Java Data Grid
    This isn't controversial...it is just WRONG and it is FUD and hype. Sounds like "unbreakable" or "$1MM for anyone who can out-perform us" sorts of tactics. Specifically: 1. For Tangosol or Gigaspaces or any clustering solution to scale, developers have to partition data so that objects don't page around the network and so that the "n-1 ACK" problem where linear scale becomes impossible doesn't occur. Under a scalable scenario, tangosol presents a "server home" for each object and is thus not strictly peer-to-peer clustering like JGroups or Tribes. Terracotta with clustered TC servers is no different. There is no way to convince the TSS community that you can simultaneously build a single system image (copy everything everywhere) and scale. Since as soon as you start partitioning Tangosol has a server too, claiming "true clustering" and simultaneously demanding partitioning for scale is just hype. Worse yet, when running partitioning data services inside the app server, it is just masked as part of the app server instance and is hard to manage and track. 2. I like the attempt to call Terracotta "circa 1995" and "database-like." We just helped a customer migrate from Oracle in under 4 weeks from initial download to full integration and production testing. They used to store transient object data, serialized in Oracle and moved that transient data to pure OO in-memory sharing on Terracotta. Why switch from Oracle to Terracotta if we are "database-like"? Simple--we are not. We are highly scalable object-oriented network attached memory that moves only memory-level object deltas on the network, only to the JVMs that have that need the data. Example: customer object where the 16 character password changes...Terracotta pushes only the 16 characters and thus out scales other solutions in most apps that have tried us. And, since Object Identity is preserved and serializable interface is not needed, none of this comes at the price of the complexity customarily present in what Cameron calls "clustering". 3. Cameron, please do not spread FUD around our license. There are many people happily both using us for free and editing our code without virality forcing them to be OSS. The reality is we are Mozilla-based which means full access to our source and full rights to edit our stuff w/o forcing your products and source to be open (no virality). The thing I think you are trying to cloud the OSS issue around is Attribution and there are OSI-approved attribution-based licenses: http://www.nabble.com/Attribution---the-Adaptive-Public-License-t3176529.html Read the definition of OSS: http://www.opensource.org/docs/osd Terracotta's license honors it, period. 4. As for our pricing, our customers have all been very happy with the price / performance of Terracotta in that it significantly lowers dev costs, production HW counts, and more than pays for itself. Tangosol was $16K per core at the beginning of this year...which is 4-6X more expensive than Terracotta's Enterprise Subscription. If you can get an ROI for customers, it stands to reason that we can given that we are cheaper. Furthermore, we have many customers who get support from our OSS community forums and aren't forced to pay one body of experts to get our stuff integrated. 5. As for Tangosol being "completely HA", this is not at all what I heard from our customers. They tend to fear the split brain / network partitioning problem not just because it is possible "it happens regularly" with clustering implementations. HA != split brain. Quotes from Cameron in Tangosol's dev forum: http://www.tangosol.net/forums/thread.jspa?threadID=356&messageID=2625 [gone since Oracle took over...sorry. But someone might be able to dig it up wherever the forums have migrated] * "The 'link breaks' problem is known as the "split brain" scenario. There is no single correct answer for the split brain, and there cannot be (by principle) a single correct answer for it." * "If the split brain were to happen, the part during resync that decides that it is not the real brain will join the rest of the brain, and in doing so will accept that the rest of the brain is managing the correct cache. * The key is to prevent a split-off portion from doing any processing "on its own" until it can verify that it is truly the "remaining few" (versus a split off piece). This is the role of a quorum approach, for example, in which no work is processed (without manual intervention) while there are fewer than a certain configured number of servers." Right now it requires programming; in 3.x it will be built in.
  30. Re: Oracle Coherence Data Grid[ Go to top ]

    Quotes from Cameron in Tangosol's dev forum:

    http://www.tangosol.net/forums/thread.jspa?threadID=356&messageID=2625 [gone since Oracle took over...sorry. But someone might be able to dig it up wherever the forums have migrated]
    No digging necessary, the forums are still at the same location http://forums.tangosol.com. Rob Misek Oracle Coherence: The Data Grid
  31. Re: Oracle Coherence Data Grid[ Go to top ]

    Ari - Your understanding of distributed computing and clustering in general seems very weak. I don't know where to even start.
    For Tangosol [..] to scale, developers have to partition data ..
    No. Developers do not have to partition data. Our partitioning has been dynamic and automatic since June, 2002, when we initially introduced the technology to the market.
    .. so that the "n-1 ACK" problem where linear scale becomes impossible doesn't occur.
    Coherence has never had to do any sort of "n-1 ACK" .. you must be thinking of jGroups.
    Under a scalable scenario, Tangosol presents a "server home" for each object and is thus not strictly peer-to-peer clustering like JGroups or Tribes.
    I don't know where to start. Yes, Coherence _is_ peer-to-peer, with a full n-way peer mesh. Just because it's not chatty doesn't mean it's not peer-to-peer. (We like to think of it as "well architected" ;-)
    terracotta with clustered TC servers is no different.
    The last I heard, terracotta used a TCP/IP connection from every client to the terracotta server, and -- on suspected failure -- can open a new TCP/IP connection to another "backup" server (which loads the data off a shared disk) if the main one goes down. That doesn't sound like a cluster. It sounds like client/server. Chutney was doing this for Java years ago ;-)
    There is no way to convince the TSS community that you can simultaneously build a single system image (copy everything everywhere) and scale.
    Single System Image does not imply full replication. It implies a Single System Image, which Coherence can provide across any of a dozen or so complex topologies. (Of course, among other things, we do support full replication as a topology.)
    Since as soon as you start partitioning Tangosol has a server too, claiming "true clustering" and simultaneously demanding partitioning for scale is just hype.
    I am quite proud to be accused of hype by you, Ari :-) However, we do true peer-to-peer clustering, and one of the data management services that we mount on our clustering is the load-balanced fault-tolerant partitioning topology. And it scales well enough to handle several stock markets and other major financial markets and many of the biggest transactional web sites in the world.
    Worse yet, when running partitioning data services inside the app server, it is just masked as part of the app server instance and is hard to manage and track.
    Masked? It's called locality of reference -- co-locating business logic and data. Our customers can _choose_ to do this, but they don't have to.
    Cameron, please do not spread FUD around our license.
    I would never knowingly spread FUD about your license. It is either OSI-approved or it is not. Please clarify for me: Is your license OSI approved (yes or no)? My understanding (please correct me if I am wrong) is that your license cannot be OSI approved.
    Tangosol was $16K per core at the beginning of this year..
    This is not correct. Our high end edition had a list price of $16k per socket, or $8k per core (customer's choice). Nonetheless, if you have a published price list, I would be glad to take back my claim and admit it was erroneous. It was based on information from a potential customer of yours who had received a quote from your company, and perhaps they had inadvertently mislead me in an attempt to get us to raise our prices. Peace, Cameron Purdy Oracle Coherence: The Java Data Grid
  32. Re: Oracle Coherence Data Grid[ Go to top ]

    Ari - Your understanding of distributed computing and clustering in general seems very weak. I don't know where to even start.
    For Tangosol [..] to scale, developers have to partition data ..
    No. Developers do not have to partition data. Our partitioning has been dynamic and automatic since June, 2002, when we initially introduced the technology to the market.
    .. so that the "n-1 ACK" problem where linear scale becomes impossible doesn't occur.
    Coherence has never had to do any sort of "n-1 ACK" .. you must be thinking of jGroups.
    Under a scalable scenario, Tangosol presents a "server home" for each object and is thus not strictly peer-to-peer clustering like JGroups or Tribes.
    I don't know where to start. Yes, Coherence _is_ peer-to-peer, with a full n-way peer mesh. Just because it's not chatty doesn't mean it's not peer-to-peer. (We like to think of it as "well architected" ;-)
    terracotta with clustered TC servers is no different.
    The last I heard, terracotta used a TCP/IP connection from every client to the terracotta server, and -- on suspected failure -- can open a new TCP/IP connection to another "backup" server (which loads the data off a shared disk) if the main one goes down. That doesn't sound like a cluster. It sounds like client/server. Chutney was doing this for Java years ago ;-)
    There is no way to convince the TSS community that you can simultaneously build a single system image (copy everything everywhere) and scale.
    Single System Image does not imply full replication. It implies a Single System Image, which Coherence can provide across any of a dozen or so complex topologies. (Of course, among other things, we do support full replication as a topology.)
    Since as soon as you start partitioning Tangosol has a server too, claiming "true clustering" and simultaneously demanding partitioning for scale is just hype.
    I am quite proud to be accused of hype by you, Ari :-) However, we do true peer-to-peer clustering, and one of the data management services that we mount on our clustering is the load-balanced fault-tolerant partitioning topology. And it scales well enough to handle several stock markets and other major financial markets and many of the biggest transactional web sites in the world.
    Worse yet, when running partitioning data services inside the app server, it is just masked as part of the app server instance and is hard to manage and track.
    Masked? It's called locality of reference -- co-locating business logic and data. Our customers can _choose_ to do this, but they don't have to.
    Cameron, please do not spread FUD around our license.
    I would never knowingly spread FUD about your license. It is either OSI-approved or it is not. Please clarify for me: Is your license OSI approved (yes or no)? My understanding (please correct me if I am wrong) is that your license cannot be OSI approved.
    Tangosol was $16K per core at the beginning of this year..
    This is not correct. Our high end edition had a list price of $16k per socket, or $8k per core (customer's choice). Nonetheless, if you have a published price list, I would be glad to take back my claim and admit it was erroneous. It was based on information from a potential customer of yours who had received a quote from your company, and perhaps they had inadvertently mislead me in an attempt to get us to raise our prices. Peace, Cameron Purdy Oracle Coherence: The Java Data Grid
  33. So good he said it twice[ Go to top ]

    Nice answer Cameron, it's good to hear there's still passion behind these technologies. You realise that I can't use Tangosol any more, before it was good technology and I felt a kinship for a fellow start-up but now you've just got good technology. :-) -John-
  34. Re: Oracle Coherence Data Grid[ Go to top ]

    Ari -

    Your understanding of distributed computing and clustering in general seems very weak. I don't know where to even start.

    Come on Cameron. Seems to me you and I are talking past each other and focusing on nuances of each of our unique experiences to make points. I want to make just a few points and not to defend myself, but because I think it will be valuable for the community. After I make these points, feel free to bash my intelligence, experience, looks, or whatever you see fit and I will try to let it all go ;) My points are: 1. Clustering is not defined as "whatever Tangosol does." from wikipedia: the [connection] of many low-cost computers to be used as one larger computer. By saying we are not clustering, you are making a value judgement on our implementation of "connection." Our user community has made a value judgement as well. They use us for HTTP session clustering, clustered caching (EHCache in particular), and coordination grids. 2. In clustering there are trade-offs to make. I get frustrated with your message because you talk about different uses of Tangosol (partitioned, replicated, near, far, grid, db caching, etc.) as if they are all the same. Paritioning, fault tolerance, linear scale, all seem to be delivered at once. This asks the reader to either be super-knowledgeable about clustering and know that, for example, some algorithms cannot be made parallel and so data partitioning strategies cannot apply. Without Locality of Reference commensurate with data partitioning, sometimes there is no linear scale. You just cannot convince me that there is no space / time trade off. Apps either write everything to disk on n+1 nodes or they blast copies around. If you write it to disk you have a SPoF. If you blast copies around, you have a network bottleneck. And no, writing to 2 nodes or even 3 is not always HA-enough (one of those nuanced uses of Tangosol that the average reader won't pick up on in this thread). 3. Terracotta is OSS by the following gernally-accepted definition: Can it be used freely? yes. Can it be changed and modified? yes. Can it be bundled? yes. Is the source available? yes. Terracotta has not gone to OSI for approval. There is a group of people working on an attribution license and we will wait before taking any action. Please, again, do not make claims like "your license cannot be OSI approved." It is simply not true. There is a license called APL approved by the OSI with a far more aggressive attribution requirement. 4. We range from $4K - $10K per node and we are generally 3 - 4X cheaper than alternatives. That said, our pricing is irrelevant because our users don't have to pay. Our mean-time-to-resolution on our OSS forum is under 2 hours. Take a look at http://forums.terracotta.org. There are about 20 new posts a day and as many new users coming online with the technology. No matter how much FUD gets spread about our pricing, our license, or our longevity, the fact is its growing by leaps and bounds all the time. Let's talk in developers' terms, not our own. Database abuse is the enemy of scale. Going outside the app tier to write to databases, message queueing, etc. purely in the name of scale and HA is the enemy. What Terracotta does is an orthogonal notion of HA and Scale to the 3 tier system. Its not about "App tier" or "database tier". anymore. Its about a purpose-built solution for the need to share information on scaled out servers.
  35. Re: Oracle Coherence Data Grid[ Go to top ]

    Ari - Thanks for the blatantly honest post :) Trust me, I don't want you to agree with me on everything. I like being in a market where we can occasionally have a fun time arguing. Besides, despite any of our best efforts to win everything, our competitive efforts are cumulatively growing the entire market, educating customers about what things they can now accomplish that were impossible before, etc. This is definitely a great market to be involved with, and no matter what you may think, I'm quite glad to see lots of competition and different approaches to solving these problems. So while I don't agree with you on most of the things you wrote (no surprise!), I am glad you don't agree, and I will only comment on one thing:
    And no, writing to 2 nodes or even 3 is not always HA-enough (one of those nuanced uses of Tangosol that the average reader won't pick up on in this thread).
    This is not something that we hide. We are very clear that we are NOT a database, but rather we are an in-memory data grid, and most of our customers use a database (e.g. Oracle RAC) behind Coherence to manage the persistent and durable system of record. With a decent data center (redundant cabling and network devices, battery backup, generator, etc.), these systems are continuously available. Even on old versions of our software (e.g. our 2.0 release), we have customers that have run for years without a second of downtime. As evidenced by the GEICO Partner Excellence Award (awarded after Coherence survived an HA database failure, among other things) and other customer references of non-stop applications despite systemic failure of critical systems, "writing to 2 nodes or even 3" is often exactly the HA they require, and typically represent the highest availability in their environments. Peace, Cameron Purdy Oracle Coherence: The Java Data Grid
  36. Re: Price list?[ Go to top ]

    I'm not seeing it on here:
    http://www.oracle.com/corporate/pricing/pricelists.html
    +1.

    This is just naive curiosity. Are there an prices yet?
    I guess you have the best technology. But there is still no answer to my question.
  37. Re: Oracle Coherence Price list?[ Go to top ]

    Hi Rex,
    I'm not seeing it on here:
    http://www.oracle.com/corporate/pricing/pricelists.html
    +1.

    This is just naive curiosity. Are there an prices yet?

    I guess you have the best technology. But there is still no answer to my question.
    The price list has been updated with the Coherence information. Please feel free to contact me (rob dot misek at oracle.com) if you have any follow up questions. Rob Misek Oracle Coherence: The Data Grid
  38. Re: Oracle Coherence Price list?[ Go to top ]

    The price list has been updated with the Coherence information.
    Thanks for your response! That's pretty cool, I can change the Oracle price list!!! (Don't ask me for more changes, I'm busy with adding java support to the iPhone ;-)
  39. Great! When free and time unlimited development licenses will be available for Enterprise Edition functionality?
  40. Great!

    When free and time unlimited development licenses will be available for Enterprise Edition functionality?
    Thanks Costin Were working on some nice surprises on that area - we'll probably announce some of them at in few weeks time. Stay tuned. Nati S. Write Once Scale Anywhere
  41. Great!

    When free and time unlimited development licenses will be available for Enterprise Edition functionality?


    Thanks Costin
    Were working on some nice surprises on that area - we'll probably announce some of them at in few weeks time.

    Stay tuned.
    Nati S.
    Write Once Scale Anywhere
    Waiting impatiently!
  42. Congratulations to the GigaSpaces team! We welcome their emphasis on Spring support, which offers an interesting option for our users. Rod Johnson, Interface21 - Spring from the source
  43. Congratulations to the GigaSpaces team! We welcome their emphasis on Spring support, which offers an interesting option for our users.

    Rod Johnson, Interface21 - Spring from the source
    Thanks Rod! Going the "full monty" with Spring was a pretty easy decision for us - I think that beyond the fact that Spring simplify the classic server side developments through a POJO and declarative driven approach, Spring provides an excellent platform for innovation and helps vendors like GigaSpaces introduce our innovative solutions to existing applications by abstracting the actual application from the underlying runtime implementation. Beyond that i think that interface21 as a company has been a great partner and helped integrating space based technologies into Spring - one of those integration point that was initially developed by Rod himself and later by Costin and was enhanced extensively with the new openspaces framework is our spring based Remoting: I would welcome anyone that is interested in further details to join our Session and BOF at the Spring ONE event in Brussle next week. Looking forward to see you at the event. Nati S Write Once Scale Anywhere