Too see how fast the MongoDB for large scale number crunching stuff, I am fiddling with MongoDB (among the other alternatives). Idea is to see how fast data can be inserted in to MongoDB.
I wrote preliminary c++/java versions and trying do optimizations in java-driver.
So I generated simple schema with 6 Strings as key-value pairs and 250 numbers in Array.
Test setup: AMD Phenom II X4 965 , 4 GB DDR3 RAM, 512 GB SATA HDD, Fedora 12 Linux
Maximum speed that I could achieve in java-version ( driver 1.4) is 1,000,000 records in 26.5 seconds -- which turns around 37,735 records/sec
Total disk size for above data is 2,489 MB -- which turns about around 93.9 MB/sec
My HDD (SATA) is having rate of 120 MB/sec -- via unix's dd-command
Original mongo-driver for java (both 1.4 and 2.0.rc.X releases) have some performance issues, after some hacking around 2 days, performance is reached to above level (3 times). Will shortly publish the changed code at mongodb-site.
Will shortly publish more code/benchmarking data :)