Case Study #1 Solr Italia Online
- Solr version : 8.8
- # of nodes: 2 clusters each 5 nodes
- hardware specs (CPU disk Servers) : 40gb memory, 12 cues , 70%
- Index size on disk 15 gb
- Query load 100 million queries per day
- Data ingestion frequency : Real time constantly
- No. of documents per collection 4 million docs / 1 collection
- What all features of solr they are using? Faceting , grouping , highlighting , autosuggest
- Technology of front end application: Python application
- Query response time before NB services : 4 seconds
- Query response time after NB services : 100m
Before:
Solution
- Switch solrconfig Solr 6.2 -> solr 8.8
- Increase soft commit to 30 secs and hard commit to 120 secs
- Grouping -> Collapsing
- Change facet.limit=-1 to 1000 for fields cd_id_sede_group_ft & ds_ls_kwd_fam_ita_ve_f
- Use use facet.method=enum for field fl_service
- Enable docvalues for field ds_ls_cat_ita
- Change replica type to TLOG for 2 and PULL for 3
- Increase soft commit to 90 secs and hard commit to 240 secs
- Nested Documents
- Increase RAM in all 10 Solr nodes from 40G to 50G
- Use G1GC instead of ConcurrentMarkSweep
- Added middle level of cache
- Increasing the replica count by double
After: 85% Reduction
Thank You
For more information about us, Contact us at:
Let's break ice
Email Us