Planned Active Use of Elastic CoD Into Power Systems Solutions
Planning to actively use Elastic Capacity on Demand (CoD) can help reduce the overall cost of your computing environment. Historically, use of Elastic CoD has been positioned for temporary events like planned growth, Live Partition Mobility operations, PowerHA failovers or disaster recovery. CoD is a premium feature offered with the Power Systems enterprise class servers.
Each of the POWER8 Enterprise servers (E850, E870, E880) offers CoD. A variety of options are available for activating CoD processor core and memory resources:
- Permanent Static – Traditional activations purchased with a server
- Permanent Mobile - For use with Power Enterprise Pools (E870/E880 only)
- Trial – Temporary no-charge activation of resources for 30 days
- Elastic – Temporary use billed in 24-hour increments
- Utility - Automatic deployment of resources to address peak use periods (cores only, billed in 5-minute increments)
Many of these features are also available with POWER7 enterprise servers. The remainder of this article will focus on evaluating the planned use of Elastic CoD vs. additional permanent activations for known workload peaks.
Elastic CoD Highlights
Elastic CoD (formerly known as On/Off CoD) allows temporary activation of processor cores and memory. Cores can be activated in increments of one and memory in 1 GB increments. Core and memory activations are independent of one another. Once proper contracts are in place with IBM, HMC administrators can activate CoD resources as needed. Usage must be reported to IBM monthly and billing is generated quarterly. One of the big cost advantages that Elastic CoD presents in total cost of ownership (TCO) studies is the no-charge inclusion of Power Systems software within the 24-hour per core usage charge. The following software products are included in the Elastic CoD charge for each 24-hour processor core usage period:
- AIX/Linux or IBM i
- IBM Cloud Manager with OpenStack
- IBM Spectrum Scale (formerly GPFS)
Note that the 24-hour core usage charge is different for AIX/Linux and IBM i. When the planned use of Elastic CoD processor cores is included in a solution design, the quantity of permanent core activations required for peak processing can be reduced. This results in reduced costs for permanent core activations, Power Systems software and associated ongoing hardware/software maintenance charges.
Considering Permanent Activation After Elastic CoD Usage
When Elastic CoD usage has occurred, you might want to decide if permanent core activations should be purchased. Following is an example where the only information provided was that 216 Elastic CoD core days were used during a one-year period. Here are two different possibilities that both lead to 216 core days used within a one-year period:
Steady Use: Activation of 1 core for 18 days per month
Peak Use: Activation of 6 cores for 3 days per month
Figure 1 shows the usage patterns. The Steady Use example might be a good candidate for permanent activation of one core since that core has a relatively high usage throughout the year. For the Peak Use pattern, it will probably be better to continue using Elastic CoD because the permanently activated cores would be idle most of the year. You’ll need to select a number of years to use when preparing a detailed TCO comparison. This example shows that the usage pattern (quantity plus duration) is important when choosing between permanent core activations and continues use of Elastic CoD. Both of the examples consume 216 core days per year, yet each will present a different TCO result when compared to permanent core activations.
Evaluating Solutions That Include CoD Use in the Design
The following is a comparison of S824 and E850 server options for a retail sales application. Retail sales applications often experience peak usage during the last six weeks of the calendar year. For this example, the following workload characteristics are present:
- For six weeks (42 days) per year 96 cores are required
- The remaining 46 weeks only require 50 percent of peak, or 48 cores
- Two options to be considered: S824 and E850
The S824 is a scale-out server model that doesn’t offer CoD. Therefore, 96 active cores must be configured to satisfy the peak requirement. This can be accomplished with four, 24-core S824 servers. Note that 48 of these cores will be idle for 46 weeks of non-peak operation. For the E850 option, two 48-core E850s satisfy the 96-core peak-processing requirement. Since the E850 features Elastic CoD, its use can be factored into the solution design. To meet the 50 percent of peak requirement, 48 permanent core activations are required. For the six-week peak period, an additional 48 cores will be activated using Elastic CoD. The final solutions to be compared are:
- (4) 24-core S824 servers, each with:
- (2) 48-core E850 servers, each with:
- 24 cores permanently activated
- 24 cores activated using Elastic CoD for six weeks per year
A TCO comparison was prepared for each option, including the following elements:
- Hardware purchase and maintenance
- Software purchase and maintenance
- Power and cooling expenses
- Rack space (16U for S824 vs. 8U for E850)
The E850 Elastic CoD usage needs to be included in the TCO. A five-year period was chosen for the TCO and the cost of Elastic CoD usage is:
$X = 24-hour per core Elastic CoD usage fee
(2 servers) * (24 cores/server) * ($X/elastic core day) * (42 days/year) * (5 years) = ($X) (10,080)
When the TCO analysis was completed, the E850 had a lower five-year TCO. A significant portion of the E850 advantage was the 50 percent reduction in software licensing and ongoing software maintenance (SWMA) expenses. When considering the use of Elastic CoD, there will often be a break-even point where permanent activation will present a lower TCO over Elastic CoD. For this example, if the E850 Elastic CoD usage is extended to 12 weeks, the S824 solution has a lower TCO.
Reviewing Your Environment
The daily usage charge for Elastic CoD with included temporary software licensing presents a new consideration when designing Power System solutions. E850 server usage modeling coupled with a TCO analysis demonstrated that an E850 solution can be a lower cost option when compared to scale-out server options. These same design principles can be used with E870 and E880 servers where increased vertical scalability is desired.
comments powered by