We handle request rates at a few thousand requests/minute with an 8 GB heap. 95th percentile response time is 200 ms. Median (cached) is 4 ms.
An oversized heap will hurt your query performance because everything stops for the huge GC. RAM is still a thousand times faster than SSD, so you want a lot of RAM available for file system buffers managed by the OS. I recommend trying an 8 GB heap with the latest version of Java 8 and the G1 collector. We have this in our solr.in.sh: SOLR_HEAP=8g # Use G1 GC -- wunder 2017-01-23 # Settings from https://wiki.apache.org/solr/ShawnHeisey GC_TUNE=" \ -XX:+UseG1GC \ -XX:+ParallelRefProcEnabled \ -XX:G1HeapRegionSize=8m \ -XX:MaxGCPauseMillis=200 \ -XX:+UseLargePages \ -XX:+AggressiveOpts \ " wunder Walter Underwood wun...@wunderwood.org http://observer.wunderwood.org/ (my blog) > On Oct 23, 2018, at 9:51 PM, Daniel Carrasco <d.carra...@i2tic.com> wrote: > > Hello, > > I've set that heap size because the solr receives a lot of queries every > second and I want to cache as much as possible. Also I'm not sure about the > number of documents in the collection, but the webpage have a lot of > products. > > About store the index data in RAM is just an expression. The data is stored > on SSD disks with XFS (faster than EXT4). > > I'll take a look to the links tomorrow at work. > > Thanks!! > Greetings!! > > > El mar., 23 oct. 2018 23:48, Shawn Heisey <apa...@elyograg.org> escribió: > >> On 10/23/2018 7:15 AM, Daniel Carrasco wrote: >>> Hello, >>> >>> Thanks for your response. >>> >>> We've already thought about that and doubled the instances. Just now for >>> every Solr instance we've 60GB of RAM (40GB configured on Solr), and a 16 >>> Cores CPU. The entire Data can be stored on RAM and will not fill the RAM >>> (of course talking about raw data, not procesed data). >> >> Why are you making the heap so large? I've set up servers that can >> handle hundreds of millions of Solr documents in a much smaller heap. A >> 40GB heap would be something you might do if you're handling billions of >> documents on one server. >> >> When you say the entire data can be stored in RAM ... are you counting >> that 40GB you gave to Solr? Because you can't count that -- that's for >> Solr, NOT the index data. >> >> The heap size should never be dictated by the amount of memory in the >> server. It should be made as large as it needs to be for the job, and >> no larger. >> >> https://wiki.apache.org/solr/SolrPerformanceProblems#RAM >> >>> About the usage, I've checked the RAM and CPU usage and are not fully >> used. >> >> What exactly are you looking at? I've had people swear that they can't >> see a problem with their systems when Solr is REALLY struggling to keep >> up with what it has been asked to do. >> >> Further down on the page I linked above is a section about asking for >> help. If you can provide the screenshot it mentions there, that would >> be helpful. Here's a direct link to that section: >> >> >> https://wiki.apache.org/solr/SolrPerformanceProblems#Asking_for_help_on_a_memory.2Fperformance_issue >> >> Thanks, >> Shawn >> >>
