GigaSpaces 5.0, grid-based application platform, released

Discussions

News: GigaSpaces 5.0, grid-based application platform, released

  1. GigaSpaces has announced the release of GigaSpaces 5.0, the latest edition of their "Space-Based Architecture" (SBA) product, which basically places services in a grid, providing dynamic scalability. SBA is geared for the distribution of stateful service-oriented transactions, according to literature from GigaSpaces.

    GigaSpaces is available in two editions, a Caching Edition and an Enterprise Edition.

    The Caching Edition provides clustering capabilities to JDBC, JavaSpaces, and JCache, as well as a management console for the clustering configuration.

    The Enterprise Edition adds JMS clustering to the Caching Edition, as well as clustering for J2EE, C++, and .NET applications, as well as dynamic deployment, configuration management, and other features.

    The new edition provides new documentation as well as early access to POJO support through Spring and Mule that was developed collaboratively with interface21, SymphonySoft and a tier-1 financial services institution in the US.

    A free 30-day evaluation version of the Caching and Enterprise versions of GigaSpaces is available for download. The Community Edition has unlimited licenses and doesn't expire, but offers limited functionality (primarily in being a JavaSpaces host, from description.)

    Threaded Messages (22)

  2. Great stuff guys, keep up the good work.

    -John-
  3. I saw a demo of their product at a local Java user's group and it was impressive. In fact, it's given me a whole new look on distributed computing. Keep up the good work guys, it's the future.
  4. Production usage[ Go to top ]

    Hi guys,

    I was wondering if people who had production experience with GS could share their experiences?

    The concept appears very interesting, just curious as how it worked out in real life?

    Thanks,

    Cedric.
  5. Production usage[ Go to top ]

    Hi guys,I was wondering if people who had production experience with GS could share their experiences?The concept appears very interesting, just curious as how it worked out in real life?Thanks,Cedric.

    I have extensive experience of GS in production, contact me off line at John dot Davies at C24 dot biz or post some general questions here if you think they'll be of use to the community.

    -John-
  6. Production usage[ Go to top ]

    I've ran Gigaspaces in production as part of a trading system for several years now. Haven't experienced a single production problem; it's rock solid.
  7. Production usage[ Go to top ]

    Hi Cedric

    In addition to John and Ming you can find below a quote from one of our recent customers who extended their use of GigaSpaces to an Enterprise level deal after seeing success with our product over a short period of time.

    "Using GigaSpaces, we are enjoying an end-to-end, on-demand, scalable solution that enables us to achieve high performance levels for our mission critical order management applications," said Julian Browne, Head of Architecture at Virgin Mobile. "GigaSpaces exceeded our performance requirements and we're looking forward to enhance this flexible, cost-effective infrastructure with the new capabilities of its latest version."

    You can find more information about our customers on the following page
    http://www.gigaspaces.com/cu_overview.html

    As you can probably imagine the list is obviously bigger then that. Unfortunately many of our customers are seeing the use of our technology as strategic and therefore avoid going publicly with such information.

    During the last quarter alone we closed several 7 figures deals (unfortunately i can't mention their names from the same reason mentioned above) Which in itself shows not just the maturity of the product but of GigaSpaces as a company and the value that people see in the technology.


    On top of that we had been able to demonstrate significant performance and scalability records in terms of performance in different large scale evaluations:

    "GigaSpaces’ Enterprise and Caching Editions have been independently and consistently benchmarked in the most demanding distributed environments, encompassing more than 2000 commodity, hyper-threaded CPUs while loading more than 1 TB of data in 20 minutes and processing 800 million events in less than 100 minutes."


    Nati Shalom
    CTO
    GigaSpaces
    "Write Once Scale Anywhere"
  8. Production usage[ Go to top ]

    Hi guys,I was wondering if people who had production experience with GS could share their experiences?The concept appears very interesting, just curious as how it worked out in real life?Thanks,Cedric.

    Cedric --

    I can give you a bunch of production references offline. Contact me at: geva at gigaspaces dot com.

    Geva Perry
    GigaSpaces Technologies
  9. Production usage[ Go to top ]

    Cedric,
    My experience in a PoC with production-level data requirements.

    Last year, I used GigaSpaces Caching edition to co-Architect and co-develop a simple version of an event correlation application. For confidentiality reasons, I can´t disclose details, however the problem is fairly generic and the resulting SBA PoC application proved both impressive performance and "near" linear scalability.

    The problem: "Given a large set of incoming events (up to 800M for example), find a small subset of them (typically between 3 and 10) which are "related". Repeat the exercise until you find all subsets of "related" events. (restriction: one event is in, at most, 1 subset)


    The approach:

    a) Use GigaSpaces Caching Edition to store event data across a distributed javaSpaces hosted (transparently) in number of physical machines.
    b) Parallelize "correlation" by creating a simple "correlation" worker program which finds one subset of "related" events.
    c) Run lots of instances of this program in parallel

    D) Test application correlation throughput and scalability at different data volumes.


    The result:
    - Correlation Throughput: up to 75K events "correlated" per second.
    - hardware "near" linear scalability as data requirements increased.
  10. Production usage[ Go to top ]

    Cedric,My experience in a PoC with production-level data requirements.Last year, I used GigaSpaces Caching edition to co-Architect and co-develop a simple version of an event correlation application.

    Sounds like a trading program.

    BTW, the KX database is probably the best at this: http://kx.com

    And if you can bring yourself to code in K or Q, your code will be extremely and likely zero-defect. It's a fascinating approach which I would love to try once.

    The system achieves untold throughput, and it was absolutely fascinating to find out exactly why it works so well.
    For confidentiality reasons, I can´t disclose details, however the problem is fairly generic and the resulting SBA PoC application proved both impressive performance and "near" linear scalability.The problem: "Given a large set of incoming events (up to 800M for example), find a small subset of them (typically between 3 and 10) which are "related". Repeat the exercise until you find all subsets of "related" events. (restriction: one event is in, at most, 1 subset)The approach: a) Use GigaSpaces Caching Edition to store event data across a distributed javaSpaces hosted (transparently) in number of physical machines.b) Parallelize "correlation" by creating a simple "correlation" worker program which finds one subset of "related" events.c) Run lots of instances of this program in parallelD) Test application correlation throughput and scalability at different data volumes. The result: - Correlation Throughput: up to 75K events "correlated" per second.- hardware "near" linear scalability as data requirements increased.

    That's the throuhgput. What was the minimum latency, i.e. what was the minium amount of time it took for a new event to generate a new correlation?

    Guglielmo

    Enjoy the Fastest Known Reliable Multicast Protocol with Total Ordering

    .. or the World's First Pure-Java Terminal Driver
  11. Production usage[ Go to top ]

    You may be interested in some real-world production experiences we've had using GigaSpaces to implement a resilient orchestration architecture for Virgin Mobile in the UK.

    The key drivers of this project were to provide reliable and resilient orchestration within an inherently unreliable and heterogeneous enterprise infrastructure.

    You can find a whitepaper that describes the project on our web-site at http://www.psjsolutions.com.

    regards

    Steve Colwill
    PSJ Solutions Ltd
  12. Production usage[ Go to top ]

    Cedric,
    Thanks for your interest. This is version 5.0 of our product and the product has been in mission-critical production environments around the world for several years. We have quite a few reference accounts and use cases we'll be happy to share with you if you're interested. Please don't hesitate to contact me at gad at gigaspaces dot com or our info channel at info at gigaspaces dot com.

    Cheers - Gad
    GigaSpaces.
  13. Another interesting article[ Go to top ]

    Have also a look at this article, which is really interesting:

    http://today.java.net/pub/a/today/2005/06/03/loose.html

    Greetings
    Stefan
  14. Another interesting article[ Go to top ]

    We've been working with GigaSpaces to introduce JavaSpaces support into Mule. This has given us a powerful combination of enterprise grid capabilities with breadth of connectivity, data routing and transformations provided by Mule. The ultimate result is that we can build Stateful SOA applications that scale up and out, something that is not address by most other ESBs.

    Ross Mason
    SymphonySoft
  15. End of Tier based computing?[ Go to top ]

    Hi Nati,

    I read your white paper "SBA and the End of Tier based Computing" which talks about eliminating tiered architectures. How many of your customers actually collapse all of their tiers into a single space? The reason to add tiers is not just for scalability. Tiers exist also for network security and organizational ownership. For example, some companies would not allow the presentation code to be just one firewall away from their DB with customer data.

    The architecture with all tiers collapsed is quite rare, no?

    Thanks,
    Peter
  16. End of Tier based computing?[ Go to top ]

    Peter,
    Allow me to add my $.02 to this argument. The point we're making is more about consolidating the tiers through a distributed shared memory bus (in our case Jini/JavaSpaces) - data, messaging and business logic processing. The data-center management aspects (security, etc.) are still very much applicable even in a consolidated environment and we offer security management down to the object-level - allowing you to manage security vertically (up and down the "tier stack") and horizontally (across the cluster). Some of our customers are 3-letter agencies that manage some of the most secure IT environments in the world - yet they recognize tier consolidation as a major boost to their IT infrastructure.

    HTH

    - Gad
    GigaSpaces, Inc.
  17. End of Tier based computing?[ Go to top ]

    Peter,Allow me to add my $.02 to this argument. The point we're making is more about consolidating the tiers through a distributed shared memory bus (in our case Jini/JavaSpaces) - data, messaging and business logic processing. The data-center management aspects (security, etc.) are still very much applicable even in a consolidated environment and we offer security management down to the object-level - allowing you to manage security vertically (up and down the "tier stack") and horizontally (across the cluster). Some of our customers are 3-letter agencies that manage some of the most secure IT environments in the world - yet they recognize tier consolidation as a major boost to their IT infrastructure. HTH - GadGigaSpaces, Inc.

    What type of approach do you use for replication? Active-active? Active-passive?

    Enjoy the Fastest Known Reliable Multicast Protocol with Total Ordering

    .. or the World's First Pure-Java Terminal Driver
  18. End of Tier based computing?[ Go to top ]

    What type of approach do you use for replication? Active-active? Active-passive?Enjoy the Fastest Known Reliable Multicast Protocol with Total Ordering.. or the World's First Pure-Java Terminal Driver

    Both - please refer to:

    http://www.gigaspaces.com/docs/manual/Replicated_Clustered_Space.htm

    Gad
    GigaSpaces, Inc.
  19. End of Tier based computing?[ Go to top ]

    Hi Nati,I read your white paper "SBA and the End of Tier based Computing" which talks about eliminating tiered architectures. How many of your customers actually collapse all of their tiers into a single space? The reason to add tiers is not just for scalability. Tiers exist also for network security and organizational ownership. For example, some companies would not allow the presentation code to be just one firewall away from their DB with customer data.The architecture with all tiers collapsed is quite rare, no?Thanks,Peter

    Hi Peter

    Since not every one here are familiar with the paper your referring to I'd like to start with a brief intro to put some context to your question.
    In Today's tiers based architecture Tiers are both physical (in most application you can point to the exact machine that runs your presentation, business, or data tiers) and logical entities. In the whitepaper - The End Of Tier Based Computing the main argument that I was making is that tiers doesn't have to be separated physically.
    There is a sense though behind the logical separation though. However once tiers becomes logical entities they can be broken down much more easily into more fine grained entities which are referred to as services .
    The second argument that I was making is that in order to achieve truly linear scalability there need to be minimal interaction between processing units of an application. In this case adding more processing units would increase the processing power of my application linearly depending on the amount of processing boxes available. The approach that I suggested was that each processing unit would have all the tiers i.e. presentation, business and data in one unit (where unit==VM). In many high throughput applications such as the one used in the financial industry where the presentation tier is not necessarily exposed to the outside world this model works pretty elegantly. In reality there are cases such as the one you mentioned in which you will not be able to achieve this level of abstraction on all tiers. This is why I suggested a five steps approach in which each step is optional. In your specific case the presentation tier can remain a separate tier but it could be a very thin tier and the business logic as well as the data will be consolidated into a together to comprise processing unit. This means that you probably won't reach 100% linear scalability but it will still get you very close to that.
    I hope that answers your question. I sometimes refer to this approach as "googalization" your tier based application. I think that it can potentially describe what I mentioned above in less words. What do you think?


    To answer your question: "How many of your customers actually collapse all of their tiers into a single space"
    For majority of our customers achieving linear scalability was the No.1 criteria in which they measured and tested our technology. Achieving linear scalability within a high throughput stateful transactional application is a much more complex to deal with.
    The paper and the concept presented in the paper was a result of years of experience we gained while working with our customers to address that goal. The paper itself already contains some references to customer case studies that led to that approach. So the answer is that majority of our customers (around 70%) are using the SBA without necessarily calling it SBA. SBA is a new term we introduced to present the different patterns we used in the past in a coherent and more complete end to end approach.
     


    Nati S.
    CTO GigaSpaces
    "Write Once Scale Anywhere"
  20. End of Tier based computing?[ Go to top ]

    Hi Nati,I read your white paper "SBA and the End of Tier based Computing" which talks about eliminating tiered architectures. How many of your customers actually collapse all of their tiers into a single space? The reason to add tiers is not just for scalability. Tiers exist also for network security and organizational ownership. For example, some companies would not allow the presentation code to be just one firewall away from their DB with customer data.The architecture with all tiers collapsed is quite rare, no?Thanks,Peter

    But these kinds of tier are largely a logical separation not a physical one. They bring a number of conveniences that can be realised without tiers.

    The Jini security model actually allows one to assert security constraints at the service and/or object levels. This permits one to separate things into logical tiers or something more granular or something less granular.

    When we're talking about separation behind firewalls, we're talking about limiting communication and therefore asserting control - i.e. physical separation. One might have several databases that are conceptually in the data tier but are managed separately and firewalled separately with separate security constraints. i.e. one can argue these databases are all in _a_ data tier but they're not in the _same_ tier.

    Things get even more interesting when one considers that each of these databases might be serving multiple and different applications from each other. At this point, tiers aren't really present anyways - what you have is some form of segmented architecture?
  21. End of Tier based computing?[ Go to top ]

    limiting communication and therefore asserting control - i.e. physical separation

    Yes, this is my point. Application level security is not sufficient for some applications (especially externally facing web sites). In these cases, physical tiers are not a luxury - they are required for network security.

    >> what you have is some form of segmented architecture

    Right, when different departments "own" different pieces of a federated application, physical separation of the application components onto different physical 'processing units' can be the reality due to budgeting, politics, etc.

    From where I sit, both of these scenarios are quite common, and so I was surprised to see the Gigaspaces white paper in which they advocate a consolidated physical model.
  22. Community Edition[ Go to top ]

    It is very difficult to see the actual features of the the community edition. The documentation of the edition is limited to a simple faq and it contains several broken links. Just as an example I cannot determine whether I can make a simple single node caching feature that is used by a number EJBs or whether I need the caching edition.
    I think you should put some effort in updating your product documentation.
    Gurli.
  23. Community Edition[ Go to top ]

    Gurli,

    Thanks for the comment about the broken links. These have been fixed.

    All production documentation available in different formats (HTML , PDF and wiki) on-line and offline.
    See:
    http://www.gigaspaces.com/os_docs.html
    http://www.gigaspaces.com/os_downloads.html

    This includes:
    Quick Start, Tutorial and Examples Guide
    Writing Your First Application step by step guide
    Enterprise Edition 5.0 Manual - HTML
    Enterprise Edition 5.0 Manual – PDF
    Javadoc
    C++ API User Guide
    C# API User Guide

    The documentation covers all product editions. There is no separate documentation for each edition.
    The Community Edition can be used as a cache for several EJBs , but we would not have high-availability and scalability options (replication , failover , partitions) . These options are part of the Caching Edition. The Enterprise Edition will provide you all Caching Edition options + sophisticated Proactive Cache options that will allow you to start Cache Instances on the fly automatically based on SLA. You may also deploy POJO into the GigaSpaces Grid – so the Enterprise Edition provide you ability to scale both at the data level and also at the business logic level.

    Please note GigaSpaces provids superior performance when running the cache embedded within the application and also when running remotly with linear scalability when running partitioned mode. We provide also a cobination of these deployemnt topologies. See product examples and docuemantion for full detials. Special optimzations exists for the JavaSpaces and Map API.

    Best Regards,
            Shay
    ----------------------------------------------------
    Shay Hassidim
    Product Manager, GigaSpaces Technologies
    Write Once Scale Anywhere