Whats a reasonable expectation for SSD performance?

thorer01

I have been making the rounds and sampling dedicated servers from a number of providers. When I get a new server, i always benchmark with dd and other scripts like bench.sh and nench to get an idea of the baseline performance for sequential reads and writes.

Here's my question:

What is considered reasonable? 250 MB\s?

If the benchmarks are around 250 MB\s i just assume that it must be an older drive, or maybe only on a SATA II connection and mark it down as the best it can do. whats the threshold that you open a ticket for investigation?

I personally believe an SSD on a SATA III port should be in the neighborhood of 400 MB\s. Is that just an unreasonable outlook?

Ishaq

Depends on the brand of the drive and the technology used, whether it's single disk or RAID array, and other factors.

thorer01

Assume no raid, just a single drive.

eva2000

Sequential read/write performance isn't everything when looking at an SSD. Response time latency matters and difference can be huge for consumer vs enterprise SSDs i.e. fsync latency for databases https://www.percona.com/blog/2018/07/18/why-consumer-ssd-reviews-are-useless-for-database-performance-use-case/

I updated my sysbench.sh script with file fsync benchmarks https://github.com/centminmod/centminmod-sysbench#sysbench-fileio-fsync

Example for 2x Samsung PM863 Enterprise SSDs in raid 1 sysbench file fsync latency response time performance on a OVH Core i7 4790K dedicated using my sysbench.sh script which outputs markdown and csv formatted results too.

fsync latency response time results are min: 0.04 ms, avg: 0.06 ms, max: 6.84 ms and 99th: 0.15 ms compared to some consumer/desktop SSDs I have seen average up to 7+ ms = 116x times slower!

./sysbench.sh file-fsync            

sysbench fileio fsync prepare
sysbench fileio --time=30 --file-num=1 --file-extra-flags= --file-total-size=4096 --file-block-size=4096 --file-fsync-all=on --file-test-mode=rndwr --file-fsync-freq=0 --file-fsync-end=0 --threads=1 --percentile=99 prepare

sysbench fileio --threads=1 --time=30 --file-num=1 --file-extra-flags= --file-total-size=4096 --file-block-size=4096 --file-fsync-all=on --file-test-mode=rndwr --file-fsync-freq=0 --file-fsync-end=0 --percentile=99 run
raw log saved: /home/sysbench/sysbench-fileio-fsync-threads-1-raw.log

sysbench 1.0.15 (using bundled LuaJIT 2.1.0-beta2)
threads: 1
Block-size 4KiB
Read/Write ratio for combined random IO test: 1.50
Using synchronous I/O mode
Doing random write test
reads/s: 0.00
writes/s: 16376.35
fsyncs/s: 16376.35
read-MiB/s: 0.00
written-MiB/s: 63.97
time: 30.0000s
min: 0.04
avg: 0.06
max: 6.84
99th: 0.15

| fileio sysbench | sysbench | threads: | Block-size | synchronous | random | reads/s: | writes/s: | fsyncs/s: | read-MiB/s: | written-MiB/s: | time: | min: | avg: | max: | 99th: |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| fileio | 1.0.15 | 1 | 4KiB | I/O | write | 0.00 | 16376.35 | 16376.35 | 0.00 | 63.97 | 30.0000s | 0.04 | 0.06 | 6.84 | 0.15 |

sysbench,sysbench,threads,Block-size,synchronous,random,reads/s,writes/s,fsyncs/s,read-MiB/s,written-MiB/s,time,min,avg,max,99th 
fileio,1.0.15,1,4KiB,I/O,write,0.00,16376.35,16376.35,0.00,63.97,30.0000s,0.04,0.06,6.84,0.15