CloudGraph Virtual File benchmark
This benchmark shows the current performance of CloudGraph's Virtual File implementation. The test indicates raw disk performance with no OS-caching (writethrough).
System specifications
Western Digital RE2 500GB 7200 RPM
eSATA-connected, single-disk on motherboard-controller (Gigabyte P55-UD6)
Intel Core i5 750 2,66 GHz (4 CPU-cores, no HT)
Windows 7 Ultimate 64-bit
8GB DDR3 RAM (OCZ, 533 MHz)
Dataset
Block size: 512 bytes
Page size: 8192 bytes
Node count: 100.000
Thread count: 1/128
______________________________________________________________
Benchmark results (raw disk performance)
______________________________________________________________
Random write (multi-threaded)
Min. throughput: 2.500 blocks/sec
Max throughput: 15.000 blocks/sec
Random write (single-threaded)
Min. throughput: 2.500 blocks/sec
Max throughput: 15.000 blocks/sec
______________________________________________________________
Random read (multi-threaded)
Min. throughput: 3.200 blocks/sec
Random read (single-threaded)
Min. throughput: 3.200 blocks/sec
______________________________________________________________
Sequential write (multi-threaded)
Min. throughput: 37.000 blocks/sec
Max throughput: 92.000 blocks/sec
Sequential write (single-threaded)
Min. throughput: 20.000 blocks/sec
Max throughput: 85.000 blocks/sec
______________________________________________________________
Sequential read (multi-threaded)
Min. throughput: 3.200 blocks/sec
Sequential read (single-threaded)
Min. throughput: 39.000 blocks/sec
______________________________________________________________
Benchmark notes
The size of one block is equivalent to a raw node.
Min. throughput represents an empty cache (cold start). Max throughput represents the database-file already in cache (running server).
Note that multi-threaded sequential read and write are not actually sequential, because multiple threads are trying to read and write simultaneously. Page size has a large effect on sequential read and write performance.