New AS SSD Benchmark
AS SSD, by Alexej Schepeljanski, is a benchmarking tool for solid state drives. It includes four synthetic tests plus final scores for read/write and combined results, and three real world copy tests.
The synthetic tests include sequential read and write operation, a 4 KB random read and write test, and the same thing using 64 threads. There is an access time test for read and write operation as well. The read access time test is executed over the full storage area of the device, while write access time test and the sequential and 4 KB random tests are limited to a 1 GB test file size. The benchmark returns three scores: one for read performance, one for write performance, and one combining both results. This is the weighting scheme:
- Overall score = seq_write * 0.15 + seq_read * 0.1 + 4k_reads * 2 + 4k_writes + 4_64thrd_writes + 4_64thrd_read * 1.5
- Read score = seq_read * 0.1 + 4k_read + 4_64thrd_read
- Write score = seq_write *0.1 + 4k_write + 4_64thrd_write
You've probably noticed that the weighting for 4 KB random reads is much higher than the other areas. This is because the benchmark author, as well as different industry sources, consider this the most important performance metric.
The practical tests consist of a copy test suite that transfers an ISO image (two large files), a software package (one folder with many small files), and a game package (one folder with small and large files). The benchmark returns the effective throughput in MB/s as well as the time required to complete this workload. We like that this test uses Windows’s copy command, which effectively shows the drives’ capability of concurrent read and write operations.
We will be using AS SSD on future SSD reviews, but of course it will complement the existing benchmarks and not replace anything with the exception of h2benchw, which is not very suitable for testing SSDs that use caching.
The benchmark also supports a compression test run, which we will be using in future SSD reviews. It allows us to analyze how a drive copes with data that is well- or poorly-compressible. SandForce-based controllers rely heavily on compression, and there are interesting results to be found when you hit them with different types of data.