Improved Storage Utilization
On average, the capacity utilization of local SSDs (inside of compute servers) is only 20% – 30% due to the need to purchase all the anticipated capacity up front with the server. This leads to purchasing more storage capacity than will ever be needed.
By using shared NVMe storage, only the actual capacity needed for each server is provisioned. Key inputs to understanding how shared NVMe storage will improve your TCO:
Average SSD capacity
of your servers
The price you paid for SSDs
Current capacity utilization of your servers
Pay As You Grow
The price of flash storage continues to decline year over year alongside the introduction of new SSD technologies (such as 3D NAND) and high density SSDs. However, using local SSDs inside of compute servers means purchasing capacity today for use in 3-5 years, making that capacity more expensive than it ought to be.
By purchasing only the shared NVMe storage that you need today, and deferring the purchasing decision for tomorrow’s capacity needs, you can benefit from lower SSD pricing and have more accurate capacity planning. To understand how this impacts your TCO, the TCO calculator will capture both your current capacity utilization and what you expect your future capacity utilization to be.
Expected growth in capacity utilization
Resiliency with Efficiency
Most applications today have replication built in to protect against server and local storage failures. In order to protect against a double failure, they maintain a 3x replica scheme which requires 3x the storage capacity for data protection.
By using shared NVMe storage with RAID data protection, the replication factor can be reduced to 2x without sacrificing resiliency against SSD device failure or server failures. With the combination of 2x replicas and RAID-5, up to 3 concurrent SSD failures can be tolerated.
Replication reduction from 3x to 2x
More Ways to Improve TCO
There are additional factors that can improve your TCO depending on your environment, although they are not assumed in the TCO model for simplicity.
If these apply to you, please contact us to see how these factors further improve your TCO.
Clustered file systems and databases use shared read / write volumes across all nodes in the cluster, which further reduces the amount of capacity required. For example, moving a 25-node cluster from local SSDs with replication to shared NVMe storage with R/W shared volumes would lead to a 25-to-1 reduction in required capacity.
There are various techniques for data reduction such as deduplication, compression, etc. However, these features impact latency and throughput and are not generally used when performance is the most important consideration.
Calculate Your TCO
To understand how shared NVMe storage can improve your TCO over using local SSDs inside of compute servers, please provide the best estimates of your current application environment. We’ll show you your immediate savings as well as the savings you will see whether you refresh your compute environment every 3 years or 5 years.