activemq-artemis/docs/user-manual/en/perf-tuning.md

14 KiB

Performance Tuning

In this chapter we'll discuss how to tune Apache ActiveMQ Artemis for optimum performance.

Tuning persistence

  • To get the best performance from Apache ActiveMQ Artemis whilst using persistent messages it is recommended that the file store is used. Apache ActiveMQ Artemis also supports JDBC persistence, but there is a performance cost when persisting to a database vs local disk.

  • Put the message journal on its own physical volume. If the disk is shared with other processes e.g. transaction co-ordinator, database or other journals which are also reading and writing from it, then this may greatly reduce performance since the disk head may be skipping all over the place between the different files. One of the advantages of an append only journal is that disk head movement is minimised - this advantage is destroyed if the disk is shared. If you're using paging or large messages make sure they're ideally put on separate volumes too.

  • Minimum number of journal files. Set journal-min-files to a number of files that would fit your average sustainable rate. This number represents the lower threshold of the journal file pool.

  • To set the upper threshold of the journal file pool. (journal-min-files being the lower threshold). Set journal-pool-files to a number that represents something near your maximum expected load. The journal will spill over the pool should it need to, but will shrink back to the upper threshold, when possible. This allows reuse of files, without taking up more disk space than required. If you see new files being created on the journal data directory too often, i.e. lots of data is being persisted, you need to increase the journal-pool-size, this way the journal would reuse more files instead of creating new data files, increasing performance

  • Journal file size. The journal file size should be aligned to the capacity of a cylinder on the disk. The default value 10MiB should be enough on most systems.

  • Use ASYNCIO journal. If using Linux, try to keep your journal type as ASYNCIO. ASYNCIO will scale better than Java NIO.

  • Tune journal-buffer-timeout. The timeout can be increased to increase throughput at the expense of latency.

  • If you're running ASYNCIO you might be able to get some better performance by increasing journal-max-io. DO NOT change this parameter if you are running NIO.

  • If you are 100% sure you don't need power failure durability guarantees, disable journal-data-sync and use NIO or MAPPED journal: you'll benefit a huge performance boost on writes with process failure durability guarantees.

Tuning JMS

There are a few areas where some tweaks can be done if you are using the JMS API

  • Disable message id. Use the setDisableMessageID() method on the MessageProducer class to disable message ids if you don't need them. This decreases the size of the message and also avoids the overhead of creating a unique ID.

  • Disable message timestamp. Use the setDisableMessageTimeStamp() method on the MessageProducer class to disable message timestamps if you don't need them.

  • Avoid ObjectMessage. ObjectMessage is convenient but it comes at a cost. The body of a ObjectMessage uses Java serialization to serialize it to bytes. The Java serialized form of even small objects is very verbose so takes up a lot of space on the wire, also Java serialization is slow compared to custom marshalling techniques. Only use ObjectMessage if you really can't use one of the other message types, i.e. if you really don't know the type of the payload until run-time.

  • Avoid AUTO_ACKNOWLEDGE. AUTO_ACKNOWLEDGE mode requires an acknowledgement to be sent from the server for each message received on the client, this means more traffic on the network. If you can, use DUPS_OK_ACKNOWLEDGE or use CLIENT_ACKNOWLEDGE or a transacted session and batch up many acknowledgements with one acknowledge/commit.

  • Avoid durable messages. By default JMS messages are durable. If you don't really need durable messages then set them to be non-durable. Durable messages incur a lot more overhead in persisting them to storage.

  • Batch many sends or acknowledgements in a single transaction. Apache ActiveMQ Artemis will only require a network round trip on the commit, not on every send or acknowledgement.

Other Tunings

There are various other places in Apache ActiveMQ Artemis where we can perform some tuning:

  • Use Asynchronous Send Acknowledgements. If you need to send durable messages non transactionally and you need a guarantee that they have reached the server by the time the call to send() returns, don't set durable messages to be sent blocking, instead use asynchronous send acknowledgements to get your acknowledgements of send back in a separate stream, see Guarantees of sends and commits for more information on this.

  • Use pre-acknowledge mode. With pre-acknowledge mode, messages are acknowledged before they are sent to the client. This reduces the amount of acknowledgement traffic on the wire. For more information on this, see Extra Acknowledge Modes.

  • Disable security. You may get a small performance boost by disabling security by setting the security-enabled parameter to false in broker.xml.

  • Disable persistence. If you don't need message persistence, turn it off altogether by setting persistence-enabled to false in broker.xml.

  • Sync transactions lazily. Setting journal-sync-transactional to false in broker.xml can give you better transactional persistent performance at the expense of some possibility of loss of transactions on failure. See Guarantees of sends and commits for more information.

  • Sync non transactional lazily. Setting journal-sync-non-transactional to false in broker.xml can give you better non-transactional persistent performance at the expense of some possibility of loss of durable messages on failure. See Guarantees of sends and commits for more information.

  • Send messages non blocking. Setting block-on-durable-send and block-on-non-durable-send to false in the jms config (if you're using JMS and JNDI) or directly on the ServerLocator. This means you don't have to wait a whole network round trip for every message sent. See Guarantees of sends and commits for more information.

  • If you have very fast consumers, you can increase consumer-window-size. This effectively disables consumer flow control.

  • Use the core API not JMS. Using the JMS API you will have slightly lower performance than using the core API, since all JMS operations need to be translated into core operations before the server can handle them. If using the core API try to use methods that take SimpleString as much as possible. SimpleString, unlike java.lang.String does not require copying before it is written to the wire, so if you re-use SimpleString instances between calls then you can avoid some unnecessary copying.

  • If using frameworks like Spring, configure destinations permanently broker side and enable cacheDestinations on the client side. See the Setting The Destination Cache for more information on this.

Tuning Transport Settings

  • TCP buffer sizes. If you have a fast network and fast machines you may get a performance boost by increasing the TCP send and receive buffer sizes. See the Configuring the Transport for more information on this.

    Note:

    Note that some operating systems like later versions of Linux include TCP auto-tuning and setting TCP buffer sizes manually can prevent auto-tune from working and actually give you worse performance!

  • Increase limit on file handles on the server. If you expect a lot of concurrent connections on your servers, or if clients are rapidly opening and closing connections, you should make sure the user running the server has permission to create sufficient file handles.

    This varies from operating system to operating system. On Linux systems you can increase the number of allowable open file handles in the file /etc/security/limits.conf e.g. add the lines

    serveruser   soft  nofile  20000
    serveruser   hard  nofile  20000
    

    This would allow up to 20000 file handles to be open by the user serveruser.

  • Use batch-delay and set direct-deliver to false for the best throughput for very small messages. Apache ActiveMQ Artemis comes with a preconfigured connector/acceptor pair (netty-throughput) in broker.xml and JMS connection factory (ThroughputConnectionFactory) in activemq-jms.xmlwhich can be used to give the very best throughput, especially for small messages. See the Configuring the Transport for more information on this.

Tuning the VM

We highly recommend you use the latest Java JVM for the best performance. We test internally using the Sun JVM, so some of these tunings won't apply to JDKs from other providers (e.g. IBM or JRockit)

  • Garbage collection. For smooth server operation we recommend using a parallel garbage collection algorithm, e.g. using the JVM argument -XX:+UseParallelOldGC on Sun JDKs.

  • Memory settings. Give as much memory as you can to the server. Apache ActiveMQ Artemis can run in low memory by using paging (described in Paging) but if it can run with all queues in RAM this will improve performance. The amount of memory you require will depend on the size and number of your queues and the size and number of your messages. Use the JVM arguments -Xms and -Xmx to set server available RAM. We recommend setting them to the same high value.

    When under periods of high load, it is likely that Artemis will be generating and destroying lots of objects. This can result in a build up of stale objects. To reduce the chance of running out of memory and causing a full GC (which may introduce pauses and unintentional behaviour), it is recommended that the max heap size (-Xmx) for the JVM is set at least to 5 x the global-max-size of the broker. As an example, in a situation where the broker is under high load and running with a global-max-size of 1GB, it is recommended the max heap size is set to 5GB.

Avoiding Anti-Patterns

  • Re-use connections / sessions / consumers / producers. Probably the most common messaging anti-pattern we see is users who create a new connection/session/producer for every message they send or every message they consume. This is a poor use of resources. These objects take time to create and may involve several network round trips. Always re-use them.

    Note:

    Some popular libraries such as the Spring JMS Template are known to use these anti-patterns. If you're using Spring JMS Template and you're getting poor performance you know why. Don't blame Apache ActiveMQ Artemis! The Spring JMS Template can only safely be used in an app server which caches JMS sessions (e.g. using JCA), and only then for sending messages. It cannot be safely be used for synchronously consuming messages, even in an app server.

  • Avoid fat messages. Verbose formats such as XML take up a lot of space on the wire and performance will suffer as result. Avoid XML in message bodies if you can.

  • Don't create temporary queues for each request. This common anti-pattern involves the temporary queue request-response pattern. With the temporary queue request-response pattern a message is sent to a target and a reply-to header is set with the address of a local temporary queue. When the recipient receives the message they process it then send back a response to the address specified in the reply-to. A common mistake made with this pattern is to create a new temporary queue on each message sent. This will drastically reduce performance. Instead the temporary queue should be re-used for many requests.

  • Don't use Message-Driven Beans for the sake of it. As soon as you start using MDBs you are greatly increasing the codepath for each message received compared to a straightforward message consumer, since a lot of extra application server code is executed. Ask yourself do you really need MDBs? Can you accomplish the same task using just a normal message consumer?

Troubleshooting

UDP not working

In certain situations UDP used on discovery may not work. Typical situations are:

  1. The nodes are behind a firewall. If your nodes are on different machines then it is possible that the firewall is blocking the multicasts. you can test this by disabling the firewall for each node or adding the appropriate rules.

  2. You are using a home network or are behind a gateway. Typically home networks will redirect any UDP traffic to the Internet Service Provider which is then either dropped by the ISP or just lost. To fix this you will need to add a route to the firewall/gateway that will redirect any multicast traffic back on to the local network instead.

  3. All the nodes are in one machine. If this is the case then it is a similar problem to point 2 and the same solution should fix it. Alternatively you could add a multicast route to the loopback interface. On linux the command would be:

    # you should run this as root
    route add -net 224.0.0.0 netmask 240.0.0.0 dev lo
    

    This will redirect any traffic directed to the 224.0.0.0 to the loopback interface. This will also work if you have no network at all. On Mac OS X, the command is slightly different:

    sudo route add 224.0.0.0 127.0.0.1 -netmask 240.0.0.0