OrientDB - Performance Tuning


In this chapter, you can get some general tips on how to optimize your application that uses OrientDB. There are three ways to increase the performance for different types of database.

  • Document Database Performance Tuning − It uses a technique that helps avoid document creation for every new document.

  • Object Database Performance Tuning − It uses the generic techniques to improve performance.

  • Distributed Configuration Tuning − It uses different methodologies to improve performance in distributed configuration.

You can achieve generic performance tuning by changing the Memory, JVM, and Remote connection settings.

Memory Settings

There are different strategies in memory setting to improve performance.

Server and Embedded Settings

These settings are valid for both Server component and the JVM where the Java application is run using OrientDB in Embedded mode, by directly using plocal.

The most important thing on tuning is assuring the memory settings are correct. What can make a real difference is the right balancing between the heap and the virtual memory used by Memory Mapping, especially on large datasets (GBs, TBs and more) where the inmemory cache structures count less than raw IO.

For example, if you can assign maximum 8GB to the Java process, it's usually better assigning small heap and large disk cache buffer (off-heap memory).

Try the following command to increase the heap memory.

java -Xmx800m -Dstorage.diskCache.bufferSize=7200 ... 

The storage.diskCache.bufferSize setting (with old "local" storage it was file.mmap.maxMemory) is in MB and tells how much memory to use for Disk Cache component. By default it is 4GB.

NOTE − If the sum of maximum heap and disk cache buffer is too high, it could cause the OS to swap with huge slowdown.

JVM Settings

JVM settings are encoded in server.sh (and server.bat) batch files. You can change them to tune the JVM according to your usage and hw/sw settings. Add the following line in server.bat file.

-server -XX:+PerfDisableSharedMem 

This setting will disable writing debug information about the JVM. In case you need to profile the JVM, just remove this setting.

Remote Connections

There are many ways to improve performance when you access the database using a remote connection.

Fetching Strategy

When you work with a remote database you have to pay attention to the fetching strategy used. By default, OrientDB client loads only the record contained in the resultset. For example, if a query returns 100 elements, but if you cross these elements from the client, then OrientDB client lazily loads the elements with one more network call to the server for each missed record.

Network Connection Pool

Each client, by default, uses only one network connection to talk with the server. Multiple threads on the same client share the same network connection pool.

When you have multiple threads, there could be a bottleneck since a lot of time is spent waiting for a free network connection. This is the reason why it is important to configure the network connection pool.

The configuration is very simple, just 2 parameters −

  • minPool − It is the initial size of the connection pool. The default value is configured as global parameters "client.channel.minPool".

  • maxPool − It is the maximum size the connection pool can reach. The default value is configured as global parameters "client.channel.maxPool".

If all the pool connections are busy, then the client thread will wait for the first free connection.

Example command of configuration by using database properties.

database = new ODatabaseDocumentTx("remote:localhost/demo"); 
database.setProperty("minPool", 2); 
database.setProperty("maxPool", 5);  

database.open("admin", "admin");

Distributed Configuration Tuning

There are many ways to improve performance on distributed configuration.

Use Transactions

Even when you update graphs, you should always work in transactions. OrientDB allows you to work outside of them. Common cases are read-only queries or massive and nonconcurrent operations can be restored in case of failure. When you run on distributed configuration, using transactions helps to reduce latency. This is because the distributed operation happens only at commit time. Distributing one big operation is much efficient than transferring small multiple operations, because of the latency.

Replication vs Sharding

OrientDB distributed configuration is set to full replication. Having multiple nodes with the same copy of database is important for scale reads. In fact, each server is independent on executing reads and queries. If you have 10 server nodes, the read throughput is 10x.

With writes, it's the opposite: having multiple nodes with full replication slows down the operations, if the replication is synchronous. In this case, sharding the database across multiple nodes allows you to scale up writes, because only a subset of nodes are involved on write. Furthermore, you could have a database bigger than one server node HD.

Scale up on Writes

If you have a slow network and you have a synchronous (default) replication, you could pay the cost of latency. In fact when OrientDB runs synchronously, it waits at least for the writeQuorum. This means that if the writeQuorum is 3, and you have 5 nodes, the coordinator server node (where the distributed operation is started) has to wait for the answer from at least 3 nodes in order to provide the answer to the client.

In order to maintain the consistency, the writeQuorum should be set to the majority. If you have 5 nodes the majority is 3. With 4 nodes, it is still 3. Setting the writeQuorum to 3 instead of 4 or 5 allows to reduce the latency cost and still maintain the consistency.

Asynchronous Replication

To speed things up, you can set up Asynchronous Replication to remove the latency bottleneck. In this case, the coordinator server node executes the operation locally and gives the answer to the client. The entire replication will be in the background. In case the quorum is not reached, the changes will be rolled back transparently.

Scale up on Reads

If you already set the writeQuorum to the majority of nodes, you can leave the readQuorum to 1 (the default). This speeds up all the reads.