An Adobe Experience Manager (AEM) Assets setup contains a number of hardware, software, and network components. Depending upon your deployment scenario, you may require specific configuration changes to hardware, software, and network components to remove performance bottlenecks.
In addition, identifying and adhering to certain hardware and software optimization guidelines helps build a sound foundation that enables your AEM Assets deployment to meet expectations around performance, scalability, and reliability.
Poor performance in AEM Assets can impact user experience around interactive performance, asset processing, download speed, and other areas.
In fact, performance optimization is a fundamental task that you perform before you establish target metrics for any project.
Here are certain key focus areas around which you discover and fix performance issues before they have an impact on users.
While AEM is supported on a number of platforms, Adobe has found the greatest support for native tools on Linux and Windows, which contributes to optimum performance and ease of implementation. Ideally, you should deploy a 64-bit operating system to meet the high memory requirements of an AEM Assets deployment. As with any AEM deployment, you should implement TarMK wherever possible. While TarMK cannot scale beyond a single author instance, it is found to perform better than MongoMK. You can add TarMK offload instances to increase the workflow processing power of your AEM Assets deployment.
To improve asset upload times, use high performance storage for the Java temp directory. On Linux and Windows, a RAM drive or SSD could be used. In cloud-based environments, an equivalent high speed storage type could be used. For example in Amazon EC2, an ephemeral drive drive can be used for the temp folder.
Assuming the server has ample memory, configure a RAM drive. On Linux, run these commands to create an 8 GB RAM drive:
mkfs -q /dev/ram1 800000 mkdir -p /mnt/aem-tmp mount /dev/ram1 /mnt/aem-tmp df -H | grep aem-tmp
In Windows OS, you would have to use a third-party driver to create a RAM drive or just use high performance storage such as SSD.
Once the high performance temp volume is ready, then set the JVM parameter -Djava.io.tmpdir. For example, you could add the JVM parameter below to the CQ_JVM_OPTS variable in the bin/start script of AEM:
Because Oracle has stopped releasing updates for Java 7 as of April 2015, Adobe recommends deploying AEM Assets on Java 8. In some cases, it has demonstrated improved performance.
You should set the following JVM parameters:
Separating the data store from the segment store is recommended for all AEM Assets users. In addition, configuring the
cacheSizeInMB parameters can help maximize performance. Set
maxCachedBinarySize to the smallest file size that can be held in the cache. Specify the size of the in-memory cache to use for the datastore within
cacheSizeInMB. Adobe recommends you set this value between 2-10 percent of the total heap size. However, load/performance testing can help determine the ideal setting.
When uploading large amounts of assets to Adobe Experience Manager, to allow for unexpected spikes in memory consumption and to prevent JVM fails with OutOfMemoryErrors, reduce the configured maximum size of the buffered image cache. Consider an example that you have a system with a maximum heap (-
Xmxparam) of 5 GB, an Oak BlobCache set at 1 GB, and document cache set at 2 GB. In this case, the buffered cache would take maximum 1.25 GB and memory, that would leave only 0.75 GB memory for unexpected spikes.
Configure the buffered cache size in the OSGi Web Console. At
https://host:port/system/console/configMgr/com.day.cq.dam.core.impl.cache.CQBufferedImageCache, set the property
cq.dam.image.cache.max.memory in bytes. For example, 1073741824 is 1 GB (1024 x 1024 x 1024 = 1 GB).
From AEM 6.1 SP1, if you’re using a
sling:osgiConfig node for configuring this property, make sure to set the data type to Long. For more details, see CQBufferedImageCache consumes heap during Asset uploads.
Implementing an S3 or Shared File Datastore can help to save disk space and increase network throughput in large-scale implementations. For more information on the pros and cons of using a shared datastore, see Assets Sizing Guide.
The following S3 Data Store configuration (
org.apache.jackrabbit.oak.plugins.blob.datastore.S3DataStore.cfg) helped Adobe extract 12.8 TB of binary large objects (BLOBs) from an existing file data store into an S3 data store at a customer site:
accessKey=<snip> secretKey=<snip> s3Bucket=<snip> s3Region=us-standard s3EndPoint=<a href="https://s3.amazonaws.com/">s3.amazonaws.com</a> connectionTimeout=120000 socketTimeout=120000 maxConnections=80 writeThreads=60 concurrentUploadsThreads=30 asyncUploadLimit=30 maxErrorRetry=1000 path=/opt/author/crx-quickstart/repository/datastore s3RenameKeys=false s3Encryption=SSE_S3 proactiveCaching=true uploadRetries=1000 migrateFailuresCount=400
Adobe recommends enabling HTTPS because many companies have firewalls that sniff HTTP traffic, which adversely impacts uploads and corrupts files. For large file uploads, ensure that users have wired connections to the network because a WiFi network becomes quickly saturated. For guidelines on identifying network bottlenecks, see Assets Sizing Guide. To assess network performance by analyzing network topology, see Assets Network Considerations.
Primarily, your network optimization strategy depends upon the amount of bandwidth available and the load on your AEM instance. Common configuration options, including firewalls or proxies can help improve network performance. Here are some key points to bear in mind:
Wherever possible, set the DAM Update Asset workflow to Transient. The setting significantly reduces the overheads required to process workflows because, in this case, workflows need not pass through the normal tracking and archival processes.
By default, the DAM Update Asset workflow is set to Transient in AEM 6.3. In this case, you can skip the following procedure.
http://localhost:4502/miscadmin on the AEM instance you want to configure.
From the navigation tree, expand Tools > Workflow > Models > dam.
Double-click DAM Update Asset.
From the floating tool panel, switch to the Page tab, and then click Page Properties.
Select Transient Workflow Click OK.
Some features do not support transient workflows. If your AEM Assets deployment requires these features, do not configure transient workflows.
In cases where transient workflows cannot be used, run workflow purging regularly to delete archived DAM Update Asset workflows to ensure system performance does not degrade.
Normally, you should run purging workflows on a weekly basis. However, in resource-intensive scenarios, such as during wide-scale asset ingestion, you can run it more frequently.
To configure workflow purging, add a new Adobe Granite Workflow Purge configuration through the OSGi console. Next, configure and schedule the workflow as part of the weekly maintenance window.
If purging runs for too long, it times out. Therefore, you should ensure that your purging jobs complete to avoid situations where purging workflows fail to complete owing to the high number of workflows.
For example, after running numerous non-transient workflows (that creates workflow instance nodes), you can run ACS AEM Commons Workflow Remover on an ad-hoc basis. It removes redundant, completed workflow instances immediately rather than waiting for the Adobe Granite Workflow Purge scheduler to run.
By default, AEM runs a maximum number of parallel jobs equal to the number of processors on the server. The problem with this setting is that during periods of heavy load, all of the processors are occupied by DAM Update Asset workflows, slowing down UI responsiveness and preventing AEM from running other processes that safeguard server performance and stability. As a good practice, set this value to half the processors that are available on the server by performing the following steps:
Setting a queue to half of the available processors is a workable solution to start with. However, you may have to increase or decrease this number to achieve maximum throughput and tune it by environment. There are separate queues for transient and non-transient workflows as well as other processes, such as external workflows. If several queues set to 50% of the processors are active simultaneously, the system can get overloaded quickly. The queues that are heavily used vary greatly across user implementations. Therefore, you may have to configure them thoughtfully for maximum efficiency without sacrificing server stability.
For high-volume of workflows or workflows that are resource intensive, such as video transcoding, you may offload DAM Update Asset workflows to a second author instance. Often, the problem with offloading is that any load that is saved by offloading the workflow processing is offset by the cost of replicating the content back and forth between instances.
As of AEM 6.2 and with a feature pack for AEM 6.1, you can perform offloading with binary-less replication. In this model, the author instances share a common datastore and only send the metadata back and forth through forward replication. While this approach works well with a shared file datastore, there can be issues with an S3 datastore. Because background writing threads can induce latency, it is possible that an asset may not have been written to the datastore before the offloading job starts.
The DAM Update Asset workflow contains a full suite of steps that are configured for tasks, such as Dynamic Media Classic PTIFF generation and InDesign Server integration. However, most users may not require several of these steps. Adobe recommends you create a custom copy of the DAM Update Asset workflow model, and remove any unnecessary steps. In this case, update the launchers for DAM Update Asset to point to the new model.
Running the DAM Update Asset workflow intensively can sharply increase the size of your file datatastore. Results of an experiment performed by Adobe have shown that the datastore size can increase by approximately 400 GB if around 5500 workflows are performed within 8 hours.
It is a temporary increase, and the datastore is restored to its original size after you run the datastore garbage collection task.
Typically, the datastore garbage collection task runs weekly along with other scheduled maintenance tasks.
If you have a limited disk space and run DAM Update Asset workflows intensively, consider scheduling the garbage collection task more frequently.
Customers use images of various sizes and formats across their website or for distribution to business partners. Because each rendition adds to the footprint of the asset in the repository, Adobe recommends using this feature judiciously. To reduce the amount of resources necessary to process and store images, you can generate these images at run time rather than as renditions during ingestion.
Many Sites customers implement an image servlet that resizes and crops images at the time they are requested, which imposes additional load on the publish instance. However, as long as these images can be cached, the challenge can be mitigated.
An alternative approach is to use Dynamic Media Classic technology to hand off image manipulation entirely. Additionally, you can deploy Brand Portal which not only takes over rendition generation responsibilities from the AEM infrastructure, but also the entire publish tier.
If you customize the DAM Update Asset workflow to generate renditions using ImageMagick, Adobe recommends you modify the policy.xml file at /etc/ImageMagick/. By default, ImageMagick uses the entire available disk space on the OS volume, and the available memory. Make the following configuration changes within the
policymap section of policy.xml to limit these resources.
<policymap> <!-- <policy domain="system" name="precision" value="6"/> --> <policy domain="resource" name="temporary-path" value="/ephemeral0/imagemagick_tmp"/> <policy domain="resource" name="memory" value="1000MiB"/> <policy domain="resource" name="map" value="1000MiB"/> <!-- <policy domain="resource" name="area" value="1gb"/> --> <policy domain="resource" name="disk" value="10000MiB"/> <!-- <policy domain="resource" name="file" value="768"/> --> <policy domain="resource" name="thread" value="1"/> <policy domain="resource" name="throttle" value="50"/> <!-- <policy domain="resource" name="time" value="3600"/> --> </policymap>
In addition, set the path of ImageMagick’s temporary folder in the configure.xml file (or by setting the environment variable
MAGIC_TEMPORARY_PATH) to a disk partition that has sufficient space and IOPS.
A mis-configuration can make your server unstable if ImageMagick uses all the available disk space. The policy changes required to process large files using ImageMagick may impact the AEM performance. For more information, see install and configure ImageMagick.
configure.xml files may be found under
/usr/lib64/ImageMagick-*/config/ instead of
/etc/ImageMagick/. See ImageMagick documentation for details on the configuration file locations.
If you are using AEM on Adobe Managed Services (AMS), reach out to Adobe Customer Care if you plan to process lots of large PSD or PSB files. Experience Manager may not process very high-resolution PSB files that are more than 30000 x 23000 pixels.
XMP writeback updates the original asset whenever metadata is modified in AEM, which results in the following:
The outcomes listed consume considerable resources. Therefore, Adobe recommends disabling XMP Writeback, if it is not required.
Importing a large amount of metadata can result in resource-intensive XMP writeback activity if run workflows flag is checked. Plan such an import during lean server usage so that performance for other users is not impacted.
When replicating assets to a large number of publish instances, for example in a Sites implementation, Adobe recommends you use chain replication. In this case, the author instance replicates to a single publish instance which in turn replicates to the other publish instances, freeing up the author instance.
Adobe does not recommend auto-activating assets. However, if necessary, Adobe recommends doing this as the final step in a workflow, usually DAM Update Asset.
Make sure you implement the latest service packs and performance-related hotfixes as they often include updates to system indexes. See Performance tuning tips | 6.x for some index optimizations that can be applied, depending on your version of AEM.
Create custom indexes for queries that you run often. For details, see methodology for analyzing slow queries and crafting custom indexes. For additional insights around query and index best practices, see Best Practices for Queries and Indexing.
Some optimizations can be done on the Oak index configurations that can help improve AEM Assets performance:
Update the LuceneIndexProvider configuration:
Update index configurations to improve reindexing time:
(AEM6.1 and 6.2 only) Update the ntBaseLucene index to improve asset delete and move performance:
Browse to /oak:index/ntBaseLucene/indexRules/nt:base/properties
Add two nt:unstructured nodes slingResource and damResolvedPath under /oak:index/ntBaseLucene/indexRules/nt:base/properties
Set the properties below on the nodes (where ordered and propertyIndex properties are of type Boolean:
On the /oak:index/ntBaseLucene node, set the property
Click Save All
Monitor the error.log to see when indexing is completed:
Reindexing completed for indexes: [/oak:index/ntBaseLucene]
You can also see that indexing is completed by refreshing the /oak:index/ntBaseLucene node in CRXDe as the reindex property would go back to false
Once indexing is completed then go back to CRXDe and set the type property to disabled on these two indexes
Click Save All
Disable Lucene Text Extraction:
If your users don’t need to be able to search the contents of assets, for example, searching the text contained in PDF documents, then you can improve index performance by disabling this feature.
When creating queries that generate large result sets, use the
guessTotal parameter to avoid heavy memory utilization when you run them.
There are two major known issues related to large files in AEM. When files reach sizes greater than 2 GB, cold standby synchronization can run into an out-of-memory situation. In some cases, it prevents standby sync from running. In other cases, it causes the primary instance to crash. This scenario applies to any file in AEM that is larger than 2GB, including content packages.
Likewise, when files reach 2GB in size while using a shared S3 datastore, it may take some time for the file to be fully persisted from the cache to the filesystem. As a result, when using binary-less replication, it is possible that the binary data may not have been persisted before replication completes. This situation can lead to issues, especially if the availability of data is important, for example in offloading scenarios.
For every AEM deployment, establish a performance testing regime that can identify and resolve bottlenecks quickly. Here are some key areas to focus on.
For all network performance concerns from the customer, perform the following tasks:
To minimize latency and achieve high throughput through efficient CPU utilization and loadsharing, monitor the performance of your AEM instance regularly. In particular:
guessTotalto optimize query performance.