Pure Performance VPS vs Cloud VPS

Estimated reading time: 3 min

What is the difference between Pure Performance VPS and Cloud VPS?

Even though we do use modern hardware (for example SSD’s only for storage, 2x10G fiber uplinks and 2x10G fiber for storage) on our Cloud VPS platform, we have some challenges to overcome. Our Cloud VPS platform has been built with maximum redundancy, maximum scalability, and maximum performance in mind, but do note it was built in that specific order of importance.

Let’s take a good look at one example of a specific choice:

We have chosen Ceph as the preferred storage solution for our Cloud VPS platform. Ceph is a distributed storage model where we have chosen for a replication setting of 3 which basically means that every single piece of data will be written to 3 disks. This is absolutely great for redundancy but a definitive hit on performance. Not only because of the fact that the data needs to be written to 3 disks but also because of the fact that Ceph has higher access times than local disks because everything needs to be accessed through the network.

In short, every choice affects every other choice, and for our main platform we believe maximum redundancy is more important than maximum performance.

Pure Performance VPS

Our Pure Performance VPS platform is a product in between our Cloud VPS platform and our (performance) dedicated servers. This platform has been built with maximum performance, maximum redundancy, and maximum scalability in this order of importance. Pure Performance, no concessions.

So what are the technical differences?

  • Higher clocked CPU’s (higher CPU frequencies) for higher computing performance (min. 3.5 GHz vs min. 2.2 GHz)
  • Local NVMe disks in ZFS mirror vdev (comparable to RAID 10) for lower disk seek times and higher disk IO and throughput
  • Minor changes in virtualization software settings with regards to CPU type to achieve maximum performance

Benchmark

For our needs we have created snench, we think this is an easy, fast, and fairly good way to benchmark performance. For more info about snench please visit its git repository. For the purposes of our benchmark today we have omitted the speed test part of the benchmark, which would add nothing to our tests here.

Disclaimer

Do note this is a snapshot. We have run these tests several times and the results were comparable but YMMV. Do your own benchmarks and compare for yourself. Benchmarking with snench is easy (one-liner) and fast!

Results

First let’s have a look at a Cloud VPS:

-------------------------------------------------
 snench.sh v2021.04.15 -- https://git.snel.com/snelcom/snench
 benchmark timestamp:     2021-04-16 13:40:18 UTC
-------------------------------------------------

Processor:    Intel Core Processor (Haswell, no TSX)
CPU cores:    4
Frequency:    2294.608 MHz
RAM:          7.8Gi
Swap:         5.0Gi
Kernel:       Linux 4.19.0-16-amd64 x86_64

Disks:
sda    100G  SSD

CPU: SHA256-hashing 500 MB
    3.232 seconds
CPU: bzip2-compressing 500 MB
    5.668 seconds
CPU: AES-encrypting 500 MB
    1.192 seconds

ioping: seek rate
    min/avg/max/mdev = 50.4 us / 438.4 us / 119.3 ms / 1.48 ms
ioping: sequential read speed
    generated 5.65 k requests in 5.00 s, 1.38 GiB, 1.13 k iops, 282.2 MiB/s
ioping: sequential write speed
    generated 948 requests in 5.00 s, 237 MiB, 189 iops, 47.4 MiB/s
ioping: random 4k read speed
    generated 11.3 k requests in 5.00 s, 44.1 MiB, 2.25 k iops, 8.81 MiB/s
ioping: random 4k write speed
    generated 1.43 k requests in 5.00 s, 5.60 MiB, 286 iops, 1.12 MiB/s

Next let’s have a look at a Pure Performance VPS:

-------------------------------------------------
 snench.sh v2021.04.15 -- https://git.snel.com/snelcom/snench
 benchmark timestamp:     2021-04-16 13:42:06 UTC
-------------------------------------------------

Processor:    Intel(R) Xeon(R) Gold 5222 CPU @ 3.80GHz
CPU cores:    4
Frequency:    3791.092 MHz
RAM:          7.8Gi
Swap:         5.0Gi
Kernel:       Linux 4.19.0-16-amd64 x86_64

Disks:
sda    100G  SSD

CPU: SHA256-hashing 500 MB
    2.200 seconds
CPU: bzip2-compressing 500 MB
    3.825 seconds
CPU: AES-encrypting 500 MB
    0.724 seconds

ioping: seek rate
    min/avg/max/mdev = 26.9 us / 44.9 us / 15.8 ms / 62.4 us
ioping: sequential read speed
    generated 52.8 k requests in 5.00 s, 12.9 GiB, 10.6 k iops, 2.58 GiB/s
ioping: sequential write speed
    generated 25.4 k requests in 5.00 s, 6.21 GiB, 5.08 k iops, 1.24 GiB/s
ioping: random 4k read speed
    generated 123.8 k requests in 5.00 s, 483.5 MiB, 24.8 k iops, 96.7 MiB/s
ioping: random 4k write speed
    generated 61.9 k requests in 5.00 s, 241.8 MiB, 12.4 k iops, 48.4 MiB/s

Comments on results

If we look at the CPU benchmarks we can clearly see that every test, whether SHA256-hashing, bzip2-compressing, or AES-encrypting, takes less time to complete on our Pure Performance VPS. If we look at the disk benchmarks the results are even more clear: the seek rate is almost 10 times as fast on the Pure Performance VPS and the disk throughput and IOPS with both sequential and random data (sequential read speed, sequential write speed, random 4k read speed, random 4k write speed) are also way faster on the Pure Performance VPS.

Conclusion

As expected the benchmark results show that both CPU performance and disk performance are better on our Pure Performance VPS platform when compared to our Cloud VPS platform.

Was this article helpful?
Dislike 0
Views: 64

Reader Interactions

Leave a Reply

Your email address will not be published. Required fields are marked *

snel-knowledgebase