How can i run benchmark using only 1.5 KB objects

Benchmarks SSD Benchmarks (using ACT)
1

Hi.

I installed ACT 6.4 in my server.

I have 2 question related with benchmark config and result.

  1. How can i run using only 1.5KB objects? Aerospike said used 1.5K blocks from ssd certification document. These devices were tested at the specified speed with a 67% read/33% write ratio of 1.5 KB objects over 24 hours. - Certifying Flash Devices (SSDs) | Aerospike Documentation

    I knew to that ACT have 2 types benchmark. act_storage, act_index. Default config of act_storage read/s :2000, write/s 1000 - 67% read, 33% write record bytes : 1.5K large-block op kbytes : 128K

    Default config of act_index read/s :2000, write/s 1000 - 67% read, 33% write Block size : 4K So, Which benchmark type should I choose between storage and index for run with only 1.5K blocks?

  2. How can get a speed(tps)?

    I can get a this result of 100x test using my SSD. after 86400 sec: reads (17278272000 total) (00: 17254429247) (01: 0023325451) (02: 0000517262) (03: 0000000039) (04: 0000000001) /dev/nvme0n1-reads (17278272000 total) (00: 17254429247) (01: 0023325451) (02: 0000517262) (03: 0000000039) (04: 0000000001) large-block-reads (203294117 total) (00: 0202334375) (01: 0000921504) (02: 0000038237) (03: 0000000001) large-block-writes (203294118 total) (00: 0203294114) (01: 0000000001) (02: 0000000002) (06: 0000000001)

    Can I refer to this result to calculate the speed(tps)?

2

I haven’t looked at ACT in a while but I believe the act_storage is what you want for validating an SSD for storing the record values (the typical use case) and the default object size for that is 1.5KiB. If the index is in memory (fairly typical) the SSD for the storage of the data is all you need.

The act_index is for the All Flash configuration, so that would be used to validate an SSD for storing the index. In such configuration the disk access is of 4KiB size and independent of the record size. Records always use 64 bytes in the index and for an All Flash configuration, one needs to have the sprigs fit in a single 4K block (64 records per sprig).

The details on how to interpret the results should be on this github repo:

Basically ./analysis/act_latency.py -l output.txt.

3

Thanks for reply, Meher.

I found anwser for my questions.

  1. 67% read/33% write ratio of 1.5 KB objects over 24 hours. That test is default configuration.

  2. speed(tps) - if i passed x100 test, that test’s speed is 300000 tps (200000read+100000write).

4

@_Jin_Park Thanks for sharing your findings.

5

This topic was automatically closed 84 days after the last reply. New replies are no longer allowed.