Unixbench

dreamcompute

#1

Hello,

Wanted to shared the Unixbench results I got from my VM:

Unixbench score: 428
More detailed benchchmarks: http://serverbear.com/benchmark/2015/07/14/FBIoqIREuceXQHMx

What I noticed about the value is the low IO and bandwidth compared to other providers.

428 is a pretty low value, compared for example with DO, similar specs are in Unixbench score of 1000. But they don’t use Ceph, so data is not as safe.

Is the low io due to Ceph?


#2

Hello Zamotcr!

I sometimes browse the forums and came across your post.

I am currently running the “Unixbench” test after looking over your scores. While the test is finishing let’s see if we can’t solve some performance issues!

First, have you created a sufficient swap file? A quick google will show you how to enable a swap file and the size it ought to be.

Second, Ceph is slow compared to dedicated fiber NAS speeds. Calling it “slow” is relative to the task it has at hand. I get 195MB/s dd from /dev/zero (That’s MegaBYTES, not Megabits)
I also get nearly 20 MB/s from Dallas, while I can buy a better link from my local fiber guys, it’s not bad for a colo at all. Keep in mind that this is a single source uploading the file during the network tests and it may or may not saturate the pipe completely.

Finally, benchmarks are highly subjective and there is not a lot of objective material you can obtain from just running a benchmark. I certainly have not felt any sort of sluggishness on my end and I compile kernels and binary’s on the regular.

I’ll be back once I get the results and we’ll figure out how to get you some better performance.


#3

Hey,

I definitely can improve the benchmark. Thanks for offering your help though.

I ran the test a second time and I got 800 this time. Considering Ceph is slower, but safer, I’m ok with it.

So I think it is fine :slight_smile:


#4

Here is my results! http://serverbear.com/benchmark/2015/07/16/yZi6AYZqElF56DLZ

Pretty good imo.


#5

Thanks everyone for posting your results and helping one another! I love that about our community.

You’re right that Ceph is going to be slower than local SSD storage or other dedicated storage options. Ceph gives us very good redundancy and reliability but the tradeoff is speed. I think speeds are pretty decent for most uses. We’re targeting desktop drive speeds and have been able to maintain that target pretty well. It does struggle with small, random writes so if you know that’s your usage pattern then you should be aware of that.