IO performance benchmark on RamNode

Posted on January 18, 2015 • 2 minutes • 283 words

Last update: 2015-01-18

I read a IO performance benchmark today comparing AWS vs DigitalOcean. It makes me curious about the IO performance on RamNode. Of course I wouldn’t expect something high since I’m running in one of the lowest tier offerred by them (It’s not even 100% SSD but just some kind of hybrid SSD-Cached). I will also ignore read test since all the VPS are container-based and read is probably heavily-cached anyway. As for the tool, I’m using fio for my IO benchmark.

Installing fio

sudo apt-get install fio
mkdir data # at ~ I suppose
nano testconfig.fio
# content of the file

[random]
rw=randread
size=4g
directory=/home/username/data
iodepth=403
direct=1
blocksize=8k
numjobs=10 # check size and your free space left (df -h)
nrfiles=1
group_reporting
ioengine=sync
loops=1

Result

Result when blocksize=8k

Run status group 0 (all jobs):
   READ: io=2048.0MB, aggrb=6485KB/s, minb=6485KB/s, maxb=6485KB/s, mint=323357msec, maxt=323357msec

Result when blocksize=16k, bw peaks at 21249.

Run status group 0 (all jobs):
   READ: io=1024.0MB, aggrb=8154KB/s, minb=8154KB/s, maxb=8154KB/s, mint=128592msec, maxt=128592msec

Result interpretation

One thing to notice is the CPU load skyrocketted. Though I only went with numjobs=2 and size=1g, the CPU load hits 3.x (300%). This is expected behavior since I’m on their Massive VPS plan . Due to this CPU issue, this is probably not the highest write speed possible as well. I would have to try it on a higher plan where CPU is irrelevant to the test. Too bad, RamNode doesn’t allow me to quickly launch an instance to test and destroy after.

So overall, DigitalOcean > AWS > RamNode. RamNode is left behind by far margin, probably due to the CPU on their lowest plan. If your application is IO-heavy, you should probably look else where.