IBM Skip to main content
Search for:   within 
      Search help  
     IBM home  |  Products & services  |  Support & downloads   |  My account

developerWorks > Java technology
JXTA 2: A high-performance, massively scalable P2P network
code170 KBe-mail it!
Learning from experience
Toward massive scalability
Configurator API improvements
A silent startup
New shell commands
Observing RPV and walker action
Observing SRDI in action
About the author
Rate this article
Related content:
Making P2P interoperable: The JXTA story
Mobile P2P messaging, Part 2
IBM Developer Kits for the Java platform (downloads)
dW newsletters
dW Subscription
(CDs and downloads)
The new JXTA release makes pragmatic changes, adapts to real-world topologies

Level: Intermediate

Sing Li ( 2: A high-performance, massively scalable P2P network)
Author, Wrox Press
11 November 2003

JXTA 2 is the second major release of the open source P2P network building substrate with a popular Java-based reference implementation. Significant design modifications have been introduced to create higher performance, massively scalable, and maintainable P2P networks. This article, which builds on Sing Li's JXTA series Making P2P interoperable, published two years ago, brings you up to date on the platform's major changes.

Since the introduction of JXTA 1.0, the open source peer-to-peer (P2P) and distributed computing communities have embraced it with a passion (see Resources for earlier developerWorks coverage of JXTA). The JXTA community flourished under a proliferation of independent open source projects aimed to test the claims made by the JXTA platform development team: that JXTA can and should become the preferred standard upon which future interoperable P2P networking applications will be built.

Learning from accumulated experience
During the last two years, the JXTA platform design team has diligently assisted the JXTA development community to better understand the basic design concepts and best practices in the use of the facilities provided by the JXTA platform API. The feedback and experience from the application development community has provided the team with valuable comments and suggestions. This feedback is the result of real-life deployment attempts, including the observation of limitations in the platform's original design, awkwardness or inconsistency in the API library, and specific pragmatic requirements that will make JXTA more applicable in building real-world P2P applications. This valuable body of accumulated experience would be impossible to gather without first introducing a workable reference implementation. As such, the first version of JXTA has served its intended purpose well.

Based on the feedback, the JXTA platform team has focused on a set of major goals in the design and implementation of JXTA 2. Table 1 shows the top design goals, along with a brief description of how the JXTA 2 implementation satisfies them (differences between JXTA 1 and JXTA 2 are highlighted where applicable).

Table 1. JXTA 2 design goals and implementations
Goal JXTA 2 Implementation
Massive scalability on today's most typical network configuration
  • JXTA 2 introduces the concept of a rendezvous super-peer network, greatly improving scalability by reducing propagation traffic (see Toward massive scalability)

  • Implementation of the shared resource distributed index (SRDI) within the rendezvous super-peer network, creating a loosely consistent, fault-resilient, distributed hash table (see Shared resource distributed index)

  • Significantly improved resource utilization, both locally on a per-node basis and network wide

  • Use of large super computers in laboratories to simulate the interactions in very large JXTA 2 networks and ensure network scalability to hundred of thousands of nodes
Increased performance
  • Improved resource allocation and reuse, both at a local per-node level and network wide

  • Local optimization, including tight control over platform memory footprint; efficiency improvement in thread handling; elimination of repeated multiple buffer copies within the protocol stack; providing control over the advertisements stored in the local cache; ability to obtain the pipe advertisement associated with a pipe without performing discovery; and more

  • Network resource improvements, including better use of available network connections by using the back-channel of a TCP connection; new TCP relay for network address translation (NAT -- see the sidebar "Network address translation device") peers; propagating only indices of advertisements instead of complete advertisements through SRDI; introduction of a bit-saving binary on-the-wire protocol format; and more

  • Local advertisement storage and indexing now performed with a customized Xindice btree database, resulting in improved performance over previous file system-based storage scheme

  • HTTP transport that now supports both 1.0 and 1.1 for more efficient use of underlying TCP connection

  • Routing hint support added to advertisements to expedite route resolution
Improved developer friendliness There are numerous API redesigns and improvements, making the APIs easier to understand and considerably more consistent. Major API changes include:
  • Message creation and handling

  • Message elements manipulation

  • Improved flexibility in asynchronous discovery handling

  • Programmatic configurator support

  • Clear, distinct separation of relay, rendezvous, router, and transport functionality

  • Unification of terminology in describing firewall, relay, and proxy elements found in current networks

  • ID factory for group ID (previously a cumbersome task to manually create)

  • JxtaSocket providing stream-based API between peers, similar to the Java platform's socket APIs

  • JxtaBiDiPipe providing bi-directional pipe capabilities for passing messages between peers
Improved reliability Changes were made on the adaptive behavior of peers to improve overall network reliability (since rendezvous and relays are the most vital services for network availability, most of the changes revolve around them):
  • Failover on rendezvous connections

  • Failover on relay connections

  • Dynamic discovery and tracking of rendezvous

  • Dynamic discovery and tracking of relays

  • Edge-peer automatically becomes rendezvous if no rendezvous were found after fixed time
Increased manageability Support for detailed remote monitoring of peers. Extensive instrumentation and metering capability is built-in with JXTA 2. This is essential for benchmarking, an essential enabler for network tuning and performance/scalability improvements. A GUI monitoring utility is also available. It marks the beginning of a long-term plan to make JXTA peers more manageable and easier to administer.
Mobile compatibility Provides proxy capabilities for JXTA implementation for Java 2 mobile edition (JXME).

Network address translation device
A network address translation device (NAT) enables multiple computers and terminals to connect to the Internet through a single network address (by translating the network address on outgoing packets and retranslating them on incoming packets). These devices are widely used by home users or business users working remotely, and are often called high speed Internet routers or Internet sharing devices.

In this article, we'll examine in more detail the key changes that make massive scalability possible in JXTA 2. We'll also do some hands-on coding with the new platform configuration API, creating a silent startup harness for the JXTA shell (a harness that you can readily reuse in your own JXTA applications). Finally, we'll explore a couple of new shell commands and take a peek at how resources are indexed and located in a JXTA 2 network.

Toward massive scalability
Network topology in the real world evolves over time. Its evolution is often influenced by external infrastructure and economic factors that typically are not in direct control of the software system designers. JXTA 1 took a pure and rather theoretical approach to using the underlying physical network topology. In contrast, JXTA 2 takes a significantly more pragmatic approach to engineering an overlay network that will work in a high performance and scalable manner on top of today's most common network topology.

JXTA 2 introduces the concept of a rendezvous super-peer network: dynamically and adaptively segregating the P2P network into populations of edge peers plus rendezvous peers. Propagation only occurs within the more stable (and smaller population of) rendezvous peers. This restriction significantly enhances scalability and reduces the possibility of network-wide message storms or flooding.

Let's take a deeper look into the implementation of this new overlay network.

Edge peer versus rendezvous super peer
Under JXTA 2, edge peers are the most transient in nature, and may come and go frequently. Most peers in a JXTA 2 network are expected to be edge peers. An edge peer does not forward query requests, and it may or may not maintain its own local advertisement cache. Most edge peers will maintain a local cache, but limited capability devices such as cell phones, PDAs, and pagers may not be able to cache, and will need to use the service of a JXME JxtaProxy (see Resources for more on JXME).

While some edge peers have direct connection to the rendezvous network, many edge peers connect to the rendezvous network through intermediate peers acting as relays or a JxtaProxy. To participate in the P2P network, an edge peer only has to know how to reach a single rendezvous. Typically, an edge peer maintains a list of known rendezvous (called seed rendezvous) and participates in dynamic discovery of new rendezvous, which enables the peer to fail-over to an alternate rendezvous and enhances overall network reliability.

Rendezvous peers are expected to be less transient and more stable (that is, once they connect to the network, they stay up and stay connected). It is anticipated that there are significantly more edge peers in a JXTA network at any time than rendezvous peers. Rendezvous peers are direct-connect peers in nature, meaning that they should not require relays or JxtaProxy to connect to other rendezvous in the network. Every rendezvous peer is a member of the rendezvous super-peer network (in the context of a particular JXTA group). Rendezvous peers forward queries that they cannot resolve based on their own cache of indices to other rendezvous; they participate in maintaining a loosely consistent distributed hash table (DHT) of all the accessible resources in the network. Every rendezvous peer maintains a dynamic view of all the known rendezvous in the network and can connect or fail-over to alternate rendezvous as network topology changes.

Every JXTA peer has the ability to assume the role of edge peer or rendezvous peer. In fact, an edge peer can adaptively become a rendezvous peer by default if it cannot connect to any rendezvous for an extended period of time. All rendezvous peers in the network can also issue queries, meaning that edge peer functionality is effectively a proper subset of rendezvous peer functionality. In general, symmetry is adaptively and pragmatically applied in a JXTA 2 network, subject to the constraint of the underlying physical network topology as well as customized user configuration.

Shared resource distributed index
The operation of the JXTA 2 network relies on its ability to resolve distributed queries (see the sidebar "The essence of P2P networking" for more on this concept). In JXTA 2, the rendezvous super-peer network forms a loosely consistent DHT to resolve distributed queries.

JXTA 2 uses a distributed algorithm, called shared resource distributed index (SRDI) to create and maintain a conceptual index of resources in the network. In JXTA, resources are described by metadata in the form of advertisements (essentially XML documents). SRDI is used to index these advertisements network wide, through a set of specified attributes. The distributed index maintained resembles a hash table, with the indexed attributes as hash keys and the hash values mapping back to the source peer containing the actual advertisement. Queries can then be made anywhere in the rendezvous network based on these attributes. In this way, SRDI can answer advertisement queries in the network by locating the peer that has the required advertisement. For example, a peer may send a query for "a pipe named LotteryService" to the network consisting of thousands of nodes. SRDI enables this query to be quickly resolved and answered by the peer that holds the "LotteryService pipe" advertisement.

Under SRDI, it is no longer necessary to remotely publish any advertisement. Only the index of the advertisements stored on a peer is published. This index information is "pushed out" to the DHT (rendezvous super-peer network) through a single connected rendezvous.

A loosely consistent DHT
The JXTA 2 network acts as an always-available, network-wide, dynamic, distributed data structure: a virtual hash table containing the index of all the published advertisements in the entire JXTA group. An edge peer can query the hash table at any time by supplying a set of attributes -- the hash keys in the table. The query is resolved by the network (actually the rendezvous super-peer network) by hashing the key to the required value (that is, the peer containing the requested advertisement), as illustrated in Figure 1:

Figure 1. Rendezvous super-peer network as a loosely consistent DHT
Figure 1. Rendezvous super-peer network as a loosely consistent DHT

In Figure 1, edge peer 1 (EP1) creates a pipe and stores its advertisement locally. The index is updated through SRDI and the DHT now knows about this advertisement. Some time later, edge peer 2 (EP2) performs a query for EP1's pipe. The rendezvous super-peer network resolves the query and notifies EP1 of the request for the advertisement. As a result, EP1 sends the requested advertisement as a response to EP2.

To create this DHT, each peer caches advertisements locally, and all locally stored advertisements are indexed. The index is pushed to the rendezvous node (JXTA 2 edge peer connects to only one rendezvous node at any time). The rendezvous super-peer network maintains the DHT containing the amalgamated indices. Queries are always sent to a rendezvous, as illustrated in Figure 2:

Figure 2. Query resolution in the rendezvous super-peer network acting as a DHT

The steps shown in Figure 2 are as follows:

  1. EP1 creates a pipe and stores its advertisement locally.
  2. The index is updated locally and the changes are pushed to the connected rendezvous (R1).
  3. Through SRDI index propagation, the rendezvous receiving the update (R1) replicates the new index information to the selected rendezvous (R3 and R5) in the super-peer network (see Rendezvous peerview and RPV walker for more information on this selection and replication process).
  4. Later, EP2 queries for EP1's pipe. This query is sent to its only connected rendezvous -- R4.
  5. R4 hashes the query's attributes and redirects the request to another rendezvous in the super-peer network -- R3.
  6. R3, having received EP1's index update through R1, immediately notifies EP1 of EP2's request.
  7. EP1 sends a response directly to EP2 containing the requested pipe advertisement. At this point, the query is resolved.
  8. EP2 may in turn decided to store the pipe advertisement, and the cycle begins again.

Thus far we've been describing the DHT maintained by the JXTA 2 rendezvous network as if it is maintained in a perfectly consistent fashion all the time. This is not currently possible because the set of rendezvous cooperating to implement the DHT may come and go (albeit less frequently than edge peers) in a P2P network. When a rendezvous goes away, the chunk of index that it is maintaining becomes unavailable for a period of time (until the responsible peers publish it again). Based on an adaptive approach, JXTA 2 can maintain a loosely consistent DHT, one that can deal with the transient nature of peers in a P2P network.

The approach that JXTA 2 takes ensures that the DHT continues to work in a suddenly partitioned network (that is, separate islands of peers created by the disappearance of a connecting peer), as well as a newly merged network (that is, a connecting peer appearing to connect formerly separate islands of peers). JXTA 2's approach involves the notion of a rendezvous peerview (RPV) and a pluggable rendezvous walker.

Rendezvous peerview and RPV walker
An RPV is maintained by each rendezvous in the super-peer network. The RPV is the list of known rendezvous to the peer, ordered by each rendezvous' unique peer ID. The hash function used in the DHT algorithm is identical in every peer and is used to determine the rendezvous that a (locally irresolvable) query request should be forwarded to.

Any rendezvous that becomes unreachable is removed from a peer's RPV. Each rendezvous in the super-peer network regularly sends a random subset of its known rendezvous to a random selection of rendezvous in its RPV. This is done to ensure eventual convergence of RPV network-wide and to adapt to any partitioning or merging occurring in the underlying physical network. Note that at any given time, the RPV maintained by different rendezvous in the network may be inconsistent with one another.

Maintaining the DHT, SRDI will store incoming index information to a selected rendezvous peer in the super-peer network based on a fixed hashing function. To cope with the loosely consistent nature, the index information is redundantly replicated to additional rendezvous peers adjacent on the RPV (recall that the RPV is an ordered list of known rendezvous by their universally unique peer ID). This will ensure that if the targeted rendezvous crashed, the probability of successfully hashing to that index information during a query is still quite high.

When resolving queries, the hashing function is performed against a rendezvous' own RPV. Because it is foreseeable that multiple existing rendezvous may have disconnected or multiple new rendezvous may have joined the super-peer network, if hashing does not immediately resolve the query, an RPV walker is introduced and forwards the query to a limited number of additional rendezvous. The algorithm used by this limited range walker is designed to be "pluggable," or customizable for more specific network scenarios. Figure 3 illustrates the redundant storage of index information.

Figure 3. SRDI-redundant storage of index information

In this figure, the hash function maps the incoming index information to R5. R1, the rendezvous that received the index information from the edge peer, will send the index information to R5, as well as replicate it to R4 and R6, increasing the availability of the index information. If, let's say, R3 received a query matching the stored index, the hash performed will send the query to R5 in R3's RPV. If, however, R5 had disappeared in the meantime, the RPV would collapse and close the gap where R5 used to reside. This means that the former R6 would become the new R5.

If R6 does not have the required index information, the topology of the network may have significantly changed. In this case, the RPV walking algorithm comes into action. The walker will walk up and down the RPV list looking for the information.

We will actually configure and work with a rendezvous super-peer network next. Before doing this, we need to take a look at some API and shell improvements in JXTA 2.

Major configurator API improvements
One major area of API improvement is in automated configuration support. It is now finally possible to use platform APIs to programmatically create a PlatformConfig file (peer advertisement) and start the JXTA platform. This means that the formerly mind-boggling array of user configurable parameters can completely be hidden.

There are multiple APIs to perform automated configuration, depending on your specific needs. Most of the configuration-related APIs are located in the net.jxta.util.config package. The net.jxta.util.config.Configurator class is a general purpose wrapper class for simplifying JXTA 2 configuration programming and is adequate for many circumstances. Under the hood, the net.jxta.util.config.Configurator class uses the lower-level classes within the same package to do the work.

In the upcoming release 2.2 of JXTA (not yet available at time of writing), these configuration APIs will be unified under an external net.jxta.ext.config package. Profiles for peers in specific roles (such as edge-peer, rendevous, relay) will further facilitate programmatic configuration.

The core objective of configuration is to create a peer advertisement that describes the peer to be started. This peer advertisement is saved by default in the .jxta directory as the PlatformConfig file.

Programming JXTA configuration for a silent startup
To facilitate our experiment, we need to start a total of six JXTA peers consisting of five rendezvous and one edge peer. To make it simple for you to duplicate this network, we would like to run all six peers on the same machine. Formerly, this would demand the Herculean task of describing all the configuration parameters to be entered manually by the JXTA GUI -- a description that could easily take the entire length of this article.

To circumvent this complexity, we'll create a starter class called This class will:

  1. Read command line parameters to configure either a rendezvous or edge peer, with unique name and transport parameters.
  2. Create the peer advertisement using the net.jxta.util.config package and store it.
  3. Start the shell with the new configuration.

Steps 1 and 2 will be performed only if no existing peer advertisement is found.

Listing 1 shows part of the code for the class. See the Resources section to download the complete code.

Listing 1. Inexpensive memory allocation in a copying collector

public class ShellStarter {
private static final String TLS_PRINCIPAL_PROP = "net.jxta.tls.principal";
    private static final String TLS_PASSWORD_PROP = "net.jxta.tls.password";
    private static final String ADDR_SEP = ":";
    private static final String PORT_PRE = "97";
    public ShellStarter() {
    public static void main(String[] args) throws Exception {
     String tpFname = 
     File tpFile = 
       new File(PlatformConfigurator.getPlatformConfigFileName());
     // only perform config if not already configured
     if (!tpFile.exists())   {          
     tcpAddress = args[1];
     rdvNode = args[3];
     int rdvNodeNum = Integer.parseInt(rdvNode);     
     myPort = Integer.parseInt(PORT_PRE + args[2] + PORT_POST);
     Vector rdvList = new Vector();
     if (rdvNodeNum < 10) 
        rdvList.add(TCP_PRE + tcpAddress + ADDR_SEP + PORT_PRE 
          + rdvNode + PORT_POST);
     pa = PlatformConfigurator.createTcpEdge(
     args[0], // peername
     "A dwPeer - " + args[0], // description
     tcpAddress, // ip
     myPort,  // ports
     rdvList    , // rdvs

     // disable multicast
      // pass in to preserve settings
     TcpConfigurator tc = new TcpConfigurator(pa);
     // enable incoming connection
     tc.setServerEnabled(true);;   // save to pa only, not file
     // configure the rendezvous
     if (isRdv) {
      // pass in to preserve rdv settings created by PlatformConfig
     RdvConfigurator rdv = new RdvConfigurator(pa);
     } // if config exists
     System.setProperty(TLS_PRINCIPAL_PROP, USER_NAME);
     System.setProperty(TLS_PASSWORD_PROP, USER_PASS);

In Listing 1, the highlighted code shows how the net.jxta.util.config.PlatformConfigurator class is used to prepare a peer advertisement that we use to start an instance of the shell silently, without invoking the GUI configurator. We first create an edge peer by calling the PlatformConfigurator.createTcpEdge() helper method with some command line arguments. However, edge peers enable multicast by default, so we want to disable this in our single machine situation. We use the net.jxta.util.config.TcpConfigurator class to turn off the multicast state. Using the same TcpConfigurator, we also enable incoming TCP connections. Finally, we check to see if the command line specifies that this should be a rendezvous node. If it does, we use a net.jxta.util.config.RdvConfigurator instance to set the peer to a rendezvous. Note also the setting of the net.jxta.tls.principal and net.jxta.tls.password system properties to bypass the login prompt.

The command line for ShellStarter takes the following parameters:

ShellStarter <peer name> <local IP or hostname> <port index> 
  <rdv port index> [edge | rdv]

Each of the edge peers and rendezvous that we create will run on the same host, but will use different TCP ports. The <port index> parameter indicates that the peer will be running at port 97?1, where "?" is the index. This will enable us to configure up to 10 total peers and rendezvous at ports 9701, 9711, 9721, and so on. For example, to create a rendezvous with peer name rdv1, on IP, on port 9711, we use the following command line:

ShellStarter rdv1 1 99 rdv

Note the use of port index 99 to indicate that this rendezvous has no other known seed rendezvous.

To create an edge peer called peer1, on the same IP, on port 9701, and seeded with the above rendezvous, we would use the following command line:

ShellStarter peer1 0 1 edge

Under the test directory, there are five startup directories for rendezvous -- rdv1 to rdv5, respectively. There is also a peer startup directory, called peer1. Each directory contains a runshell.bat file that has the corresponding parameterization for ShellStarter. You will need to edit these files to modify the IP address. Figure 4 shows the configuration of this network.

Figure 4. Staging a JXTA 2 network for experimentation

New shell commands in JXTA 2
There are several new commands in the JXTA 2 shell distribution. Among the most useful for the developer is the kdb command. You can use this command to turn on and off debugging logs for the various JXTA components while the platform is running.

The kdb command is menu driven, and can be used to set any of the 16 components to the various log priority levels (thus generating varying amounts of debugging information).

The new route command can be used to display or manipulate JXTA route table information.

Another useful command in the JXTA 2 shell, especially to facilitate the understanding of how the rendezvous super-peer network operates, is the new rdv command. This command has many options, several of which are explored in detail during our experimentation.

To start the system, first start all the rendezvous in order, from one to five. Be sure to wait for each one to completely start before starting the next one, then start peer1.

Observing RPV and walker action
Using the rdv command (see the sidebar "New shell commands in JXTA 2"), it is possible to see the RPV maintained by any of the rendezvous.

rdv -rpv

As an example, the following snippet shows the result from our rdv1:


The RPV ordering reflects the peers' JXTA peer ID ordering. We are now ready to see RPV walker in action. The rdv command has an option to run a string indexing service (for test and diagnostics only). To start this service, run the following command on each of the six peers:

rdv -start

Now, create a string on one of the rendezvous. In our case, we create "treasure" on rdv4:

JXTA> rdv -add treasure

You can use the following list option to confirm that rdv4 now contains this string:

JXTA> rdv -list treasure

Now let's see RPV walker in action. On peer1, search for the following string:

JXTA>rdv -search treasure
Sending test message
rdv has sent search query for treasure
JXTA>rdv received from : jxta://uuid-59616261646162614A78746150325
                        found: treasure

The treasure string was found immediately on rdv4. You can verify that the peer ID is indeed for rdv4 by typing whoami on rdv4.

Looking at the rdv4 shell window, you'll also see that indeed the reply is sent directly from rdv4:

Replying search query= treasure
send reply  sent

On some of the other rendezvous, you will see evidence of forwarding of the query:

Forwarding search query= treasure

Here's what happened: when you query for "treasure," the rendezvous walker is invoked to "walk" the rendezvous super-peer network. As a rendezvous client, the request is passed to the only connected rendezvous -- rdv1. It is the "walk" commencing from rdv1 that causes the propagation of the query to occur. Every rendezvous that does not have the "treasure" string forwards the request (up to an arbitrarily set time-to-live/hop-count limit of 10, and subject to loop detection) to other known rendezvous points through the walker, until the request reaches rdv4. Upon receiving the query, rdv4 replies directly to peer1.

Observing SRDI in action
Using the rdv shell command, you can easily experiment with the RPV and rendezvous walking mechanism. To actually see SRDI in action, we can set the LOG level of the SRDI message-logging mechanism to DEBUG and then cause some SRDI activity (that is, create an advertisement on a peer).

To avoid spurious messages from the diagnostic string indexing service, first turn off the services. Turn off service from rdv1 to rdv5, as well as peer1:

rdv -stop

Then on each peer, turn the SRDI messages LOG level to DEBUG using the kdb command:


KDB Main Menu

1   Change LOG configuration
q   Quit

LOG Menu

1   Global          2   EndpointRouter
3   Endpoint        4   HTTP
5   TCP             6   TLS
7   Rendezvous      8   Discovery
9   Resolver        10  Pipe
11  Relay           12  Messengers
13  Messages        14  Quota Listener
15  SRDI            16  CM

q   Quit
LOG> 15
Level [w,i,f,d,e,q or ?])> d

LOG Menu

1   Global          2   EndpointRouter
3   Endpoint        4   HTTP
5   TCP             6   TLS
7   Rendezvous      8   Discovery
9   Resolver        10  Pipe
11  Relay           12  Messengers
13  Messages        14  Quota Listener
15  SRDI            16  CM

q   Quit

Now, on peer1, we will create an advertisement that will cause SRDI index information to be pushed to rdv1 and subsequently replicated in the DHT. The easiest way to do this is simply to create a new peergroup (actually a peergroup advertisement):

newpgrp -n supergroup

If you then look at the messages coming out on rdv1, you may see SRDI messages similar to the following:

<DEBUG 14:36:07,078 Srdi:521> Pushing deltas in group NetPeerGroup
<DEBUG 14:36:07,078 Srdi:494> waiting 30000ms before sending 
deltas in group NetPeerGroup

Other rendezvous may also receive additional SRDI messages sent by rdv1 as the hashing and redundant index replication occurs.

The impact of P2P networking on everyday computing is significant, and a very large population of computing users is affected. You can't hope to design a viable P2P application creation substrate in total isolation. Consideration for user feedback, continuous requirement analysis, rethinking the design, and design iteration are all important ingredients in creating a usable substrate. JXTA is loyal to this philosophy. In its second generation, JXTA has evolved and adapted to the realistic requirements of its constituents of early adopters.

By pragmatically segregating the overlay network into an edge-peers population interconnected through a core rendezvous-peers population, JXTA 2 is significantly more deployable over today's common network topology. Problems with message broadcast storms and flooding, common in a "flat" architecture, are also alleviated.

JXTA 2's core rendezvous-peers network implements a loosely consistent DHT. All of the resources available over the P2P network are indexed and dynamically retrievable with the help of this DHT. The index maintained by the DHT is the SRDI. In contrast to the first version, advertisements are no longer propagated throughout the network; instead, only index information is propagated on demand. This approach significantly improves the use of the available network bandwidth, in a trade-off for slightly more computation during query resolution.

Another area of growing maturity is the JXTA platform API. The availability of a fully functional peer configuration API enables developers to create applications that will start silently without booting JXTA's default (and often confusing) peer configurator. We applied this API when creating our silent shell startup scripts for experimentation. Using the new shell command rdv, we were able to observe the rendezvous super-peer network in action. SRDI messaging and operations were also observed by setting DEBUG log levels through the new kdb shell command.

The goal of creating a scalable, high performance, but (most important) usable P2P network needs to be pursued relentlessly. JXTA 2 is an important step in the right direction.


  • Download the source code for this article.

  • See the series Making P2P interoperable, also by Sing Li, for coverage of the first release of the JXTA platform:
    • "Part 1, The JXTA story" provides an overview of Project JXTA and how it enables and facilitates the simple fabrication of P2P applications without imposing unnecessary policies or enforcing specific application operational models.
    • "Part 2, The JXTA command shell" gives you a hands-on tour of the JXTA shell. You'll explore its command set and extend its capabilities by writing your own custom commands using the Java programming language.
    • "Part 3, Creating JXTA systems" showcases JXTA's extension of the TCP/IP network and demonstrates that JXTA isn't bound by the constraints typical of client-server networks

  • Visit the official JXTA community to find the latest specifications, documentation, source, and binaries.

  • The white paper "Project JXTA 2.0 Super-Peer Virtual Network," by Bernard Traversat et al (Project JXTA, May 2003), describes the inner workings of the rendezvous super-peer network in intricate details.

  • See how to deploy JXME and turn mobile devices into JXTA and Jabber clients in "Mobile P2P messaging, Part 2: Develop mobile extensions to generic P2P networks" by Michael Juntao Yuan (developerWorks, January 2003).

  • For more JXME coverage, check out mobile devices in "Tips & tricks: JXTA" by Roman Vichr (developerWorks, April 2002).

  • Anne Zieger examines JXTA in her "Peer-to-peer communications using XML" article (developerWorks, April 2002).

  • Todd Sundsted's developerWorks series on "The practice of peer-to-peer computing" (March 2001 - January 2002) provides background reading on fundamental P2P computing principles.

  • Find hundreds of articles about every aspect of Java programming in the developerWorks Java technology zone.

About the author
Photo of Sing LiSing Li is the author of Professional Apache Tomcat, Early Adopter JXTA,, and Professional Jini, as well as numerous other books with Wrox Press. He is a regular contributor to technical magazines and is an active evangelist of the P2P evolution. Sing is a consultant and freelance writer and can be reached at

code170 KBe-mail it!

What do you think of this document?
Killer! (5) Good stuff (4) So-so; not bad (3) Needs work (2) Lame! (1)


developerWorks > Java technology
  About IBM  |  Privacy  |  Terms of use  |  Contact