Instaclustr has recently released its managed Cassandra on SoftLayer bare metal servers. Running on bare metal provides the fastest available Cassandra performance and great value for money.
Instaclustr has recently brought our managed Cassandra service to the IBM SoftLayer platform. One of the key drivers behind us building support for SoftLayer was allowing us to offer Cassandra on bare metal servers. We expected that bare metal would offer exceptional performance and our recent benchmarking has demonstrated that this is the case.
For this benchmarking, we followed the same approach that we took for our recent benchmarking of Cassandra 2.1 on AWS. That is, we ran Cassandra-stress against the standard install provisioned by our system with no further tuning. We adjust the number of threads used by Cassandra-stress to roughly maximise the number of operations per second achieved.
The following tables summarise the results of this testing on a 3 node Xeon_1270 cluster (the smallest SoftLayer size offered by Instaclustr) and compare these to the Cassandra 2.1 results we achieved with an i2.2xlarge on AWS (the largest and highest performing AWS size offered by Instaclustr):
|Provider||Cassandra Version||Mixed Load|
(3:1 read: write)
|Read Only||Insert Only|
|Provider||Cassandra Version||Read Test Latency (ms)|
|Median||95th Percentile||99.9th Percentile|
As can be seen from these results, bare metal servers and SSDs offer fantastic performance in terms of both high throughput and low latencies. SoftLayer bare metal is also extremely good value for money on a cost per unit of performance basis (sign up for access to our console to compare the pricing details).
It’s fair to say that we were pleasantly surprised by both the fantastic performance and value for money of the SoftLayer bare metal offerings. While AWS still offers great flexibility and a very broad set of services, SoftLayer is well worth considering, particularly if you need the highest levels of performance.