AEM Performance Optimization Tips

AEM Performance Optimization Tips

• 10% of the project efforts should be allocated to performance.
• Performance optimization is ongoing activity and should have more preference after the go live where the real load comes into the picture
• Simulate real scenarios
• Establish solid goals
• Take one bottle neck at a time
• Guidelines 
• 70% of the pages should take 100ms
• 25% of the pages should take 100-300ms
• 4% pages should take 300-500ms
• 1% of the page should take 500-1000ms
• No page should take more than 1000ms
• Focus on dispatcher caching efficiency
•  use of queries
• No of child node should not exceed 1000 nodes
• use java profiler
• use performance monitor
• collect as much information as possible before making optimization decisions
• try to replicate
• observe changes in code
• check if the issue occurs only during certain time
• monitor search request timing
• limit the concurrent workflow processing
• configure maximum parallel jobs
• Do error scenario tests
• Endurance tests
• Dispatcher Tuning 
• Calculate the dispatcher cache ratio : Pages served from cache vs page served from publisher
• Consistent page encoding
• Avoid url parameters because pages with parameters do not get cached. Use selectors instead.
• Put title image file in the same location as page so that it can be invalidated when the page is updated.
• Same way invalidate the image files used as navigation texts
• Avoid free customization of pages and provide say a definite number of customized pages so that they can be cached.
• Use sticky connections if possible.
• Read here for backup performance in detail
• set queue.maxparellel to 50% of your CPU capacity
• create indexes for the queries most frequently searched
• set -Doak.queryLimitMemory
• set -Doa.queryLimitReads 
• set -Dupdate.Limit
• set -Doak.fastQuerySize = true
• For Lucene Indexes set enable copyOnRead, copyOnWrite, Prefetch Index Files
• While using file data store set the value of cacheSizeInMB and maxCacheBinarySize
• Use offline compaction for TarMK
• Allow all css and js to combine into clientlib.css and clientlib.js respectively
• minify the files
• Use MongoMK on author only and only if high load of concurrent users expected. E.g. 1000 users or more
•