“Disaggregation” and “Composition” are Key to Cost Savings
Software Composable Infrastructure can have a significant impact on reducing data center costs, from initial deployments through day-to-day operations and lifecycle upgrades. Two key attributes of SCI provide the opportunity to reduce both capital and operating costs: 1) Disaggregation and 2) Software Composition.
Disaggregation of compute and storage resources gives you the ability to purchase and scale them separately. Software composition makes it quick and easy to combine them in any ratio, and adjust that ratio on-the-fly as needed. Together, these aspects of SCI can reduce costs in several ways:
Disk-lite servers and JBODs cost less than the same servers with internal disks. With SCI, you have the flexibility to purchase compute and storage resources from different vendors, giving you even more options. For example, additional savings can be realized by purchasing drives directly from the manufacturer instead of from the server vendor who marks up the price. In many cases, low cost 3.5” disk drives can be used instead of expensive 2.5” drives because the limitations on the number of drives that can fit in a server box are removed.
The concern that the compute-to-storage ratio is fixed for a given cluster can push administrators to over-provision processors. This could be a reasonable strategy if more and more jobs are added to a cluster as it matures, but it often results in clusters where the processors are running at low utilization levels. With DriveScale, you can configure compute and storage as needed without over-provisioning, and easily add resources later if necessary.
One of the biggest savings that comes from implementing DriveScale is the ability to upgrade compute and storage separately. Upgrading the diskless servers without having to also replace the disk drives presents huge savings in money, time and effort. Compute intensive applications will benefit from more frequent upgrades as new high-performance processor technologies become available. Upgrading disk drives can be postponed until the drives are nearing EOL, or as newer and higher capacity drives are needed.
If more compute power is needed, add more processors and spread the existing drives across them. If more storage is needed, add more disks and attach them to exiting nodes. DriveScale makes it easy to expand only the constrained resource without paying for other resources you don’t need.
In traditional scale-out architectures, once a cluster starts out with a particular disk-to-compute configuration, all additions to that cluster must conform to the same configuration. There can’t be 8 drives per node in half the cluster and 12 drives per node in the other half because the distribution of the workload among the servers would become chaotic and unpredictable. The only way to address an incorrect and wasteful configuration is to swap out the entire cluster. DriveScale lets you easily reconfigure the entire cluster at any time in response to changing requirements.
Multiple application clusters have traditionally been silo’ed in their own separate racks and underutilized resources were locked up in those silos. With DriveScale, you can quickly and easily make adjustments, moving resources between clusters so every workload operates at maximum efficiency. Silos are a thing of the past!
Diskless servers and JBODs take up about half the rack space of equivalent servers with internal drives. Without the disks blocking the airflow, compute servers can be more tightly configured with up to 8 servers in a 2u box. Since disk drives require little cooling, upwards of 100 3.5” drives can be accommodated in a 4u JBOD. Reduced rack and floor space also saves on cooling and power.
In traditional scale-out architectures, decommissioning a cluster has no assurance that a new cluster could use the configuration of the decommissioned nodes. With DriveScale, decommissioned clusters release their hardware components to pools of available resources that can easily be re-composed into completely new configurations as needed.
Since a DriveScale user can build many types of server nodes from one compute SKU and one JBOD SKU, the number of SKUs required to build out the various clusters in a data center can be drastically reduced, while still optimizing resources for each workload. This makes ordering, sparing, managing and inventory tasks easier. Spare inventory levels can be much lower.
DriveScale also helps IT respond faster to changing business requirements, recover more quickly from failures and meet or exceed demanding SLAs. See Respond Faster.