Total Pageviews

Thursday, April 11, 2013

IBM System z: The Lowest-Cost Database Server Solution

Enterprises that trusted in the mainframe myths and moved their corporate databases to distributed platforms are spending 100 percent more than necessary on database servers, creating data integrity issues and increasing the data risk exposure by constraining the ever-shrinking backup window. By moving databases from shared-nothing distributed data servers to the shared storage environment of IBM zEnterprise systems and putting the applications on Integrated Facility for Linux (IFL) processors and an IBM zEnterprise BladeCenter Extension (zBX), IT executives can reduce their ecosystem costs by more than 50 percent per year. At Robert Frances Group (RFG), we completed a Total Cost of Ownership (TCO) analysis of the traditional distributed Linux and Microsoft Windows environment vs. a zEnterprise with zBX environment that consolidates the databases on the mainframe and found the distributed environment to be twice as expensive. Our study used the standard three-year zEnterprise leasing and refresh strategy and traditional five-year purchase plan for the distributed x86 scale-out scenario. IT executives should evaluate the shared zEnterprise database server alternative to lower costs, improve productivity and reduce data risk. Additionally, IT executives should work with IBM or a third-party lessor to structure a package that best meets current and future business, financial and IT objectives. We had several ideas and hypotheses in mind before conducting our analysis. Specifically:
• The scale-out distributed server model using shared-nothing databases is costly and inefficient, creates data integrity and operational exposures, and fails as a best practice. A switch to using the mainframe as a database server eliminates the need for database duplication and synchronization since the mainframe uses a shared-everything architecture. While the acquisition cost of the zEnterprise and zBX servers collectively runs more than distributed x86-based servers, this is more than compensated for by the drastic reduction in database arrays and their associated costs. IT executives should assess the platform options holistically rather than piecemeal to identify the optimal solution.
• A zEnterprise environment can place Linux applications on IFLs and Windows applications on a zBX. Using this tightly knit, workload-optimized solution reduces the number of processors required, improves application and system management, and uses a high-speed interconnect so performance isn’t diminished when shifting to a shared-everything database engine. A zEnterprise solution enables enterprises to improve automation, control, security and visibility to their applications and databases without degrading performance. IT executives should determine which applications and databases should move to a zEnterprise environment and perform a TCO analysis to gain executive buy-in for the shift.
• Several non-financial gains accrue when moving to a shared-everything storage environment and these should also be factored into the decision-making process. Having a single copy of data means there’s only one version of the truth, all outputs and reports will be consistent and keeping things in synch won’t require manual manipulation, which is error-prone. Most enterprises today spend between 25 and 45 percent of their time synchronizing the many database copies. The associated time consumption used for duplication also creates a backup exposure; some backups don’t occur when administrators are pressed for time. Business and IT executives should consider these data integrity and risk exposures.
• Most IT executives have blindly accepted as fact the theorem that distributed processing is the least expensive solution. This hypothesis has gained ground because of a focus on a Total Cost of Acquisition (TCA) perspective. If the only valid cost analysis is the TCA of servers, then this might hold water. However, when the entire ecosystem is analyzed—including administrator costs, application and middleware software license and maintenance fees, cabling, networking, servers, storage, floor space and power and cooling—this theory falls apart.
• When the zEnterprise is used as a database server and IFLs and zBX are fully leveraged—and the analysis occurs holistically—a different picture emerges. The zEnterprise environment costs more than 50 percent less than that of a distributed x86 ecosystem, mostly due to the savings on storage, administrator, warranty and software costs.
• The mainframe architecture supports shared-everything storage while all distributed operating system platforms use a shared-nothing architecture. The mainframe architecture is unique in that multiple workloads share processors, cache, memory, I/O and storage. Moreover, zEnterprise systems provide data, IT and storage management practices and processes that facilitate and simplify the centralized, shared environment and enable application and database multitenancy. This means mainframe applications can share a single instance of a database, such as customer data, while distributed systems force the creation of a copy for each application’s use.
• Often, companies have between seven and 50 copies of the same database in use, so every terabyte of data stored is expanded by requirements for archiving, backup, mirroring, snapshots, test systems and more (see Figure 1). This data store expansion is then duplicated by the number of copies the distributed systems require. Thus, 1 TB of data in a distributed environment could grow to in excess of 100 TBs—more than 10 times the amount needed when databases are shared using a zEnterprise. There are software clustering solutions to get around this distributed duplication phenomenon and some of the storage sprawls, but they’re partial fixes and only address certain data sets.
• Mainframe storage capacity requirements are a fraction of what’s required for distributed systems. Annual acquisition costs for additional storage on a mainframe will be far less than that for distributed storage solutions. The capital expenditure (CAPEX) savings from the differential in storage costs when mainframes are used as a database engine far exceed the added expense of the mainframe hardware. The mainframe’s smaller storage footprint will reduce the operational expenditures (OPEX) and lower the TCO.

The Methodology

We hypothesized that a large Small to Midsized Business (SMB) with revenues between $750 million and $1 billion might operate a more economical data center environment if it used the new zEnterprise architecture and the mainframe as a database server. Most SMBs run their applications on Windows and/or Linux on x86-architected servers that don’t offer the advantages of a scale-up architecture. Let’s assume AB Co. (ABCo) runs 500 applications with 75 percent of them (375) executing on top of the Windows operating system. The remaining applications (125) run on Red Hat's Enterprise Linux. Additionally, 10 percent are CPU-intensive and require their own blade servers. All other applications operate under either VMware or KVM, depending on whether they’re Windows or Linux applications, respectively. The application workload growth rate is at 20 percent per year.

We also assumed a Storwize V7000 Unified Storage System houses the databases for the mainframe and distributed environments. To keep the analysis from becoming too complex, only two sizes of databases are used (1 and 2 TBs) and each application accesses 10 databases, half of each size. The storage growth rate is 25 percent. There are a total of 70 unique databases, half of each size. For the purposes of the study, only the production servers and storage are included; excluded are the archive, backup and snapshot copies of data. Because a Storwize storage solution is used, we assume a 60 percent utilization is achieved in all environments.

We further assumed that 126 TB of storage is required to handle storage needs for the first 12 months of operation. This includes an additional 20 percent for duplicate databases for the mission-critical applications. On the x86 side, since this is a shared-nothing framework, a minimum set of seven copies of databases would be needed. This results in the total initial storage capacity of 770 TB being required to support the storage needs of the first year’s operation. Finally, note that DB2 10 for z/OS is the database software used to access all databases.

The x86 server scenario uses all IBM 16-core HX5 blade servers for application and database processing. The zEnterprise uses the Central Processing (CP) environment to handle all the database interactions, exploits IFLs for all of the Linux workloads and the zBX for the Windows workloads. In this way, each workload is allocated to the server platform best-suited to perform the task. We further assumed the x86 servers were purchased and kept in operation throughout the five-year analysis period while the zEnterprise boxes were leased and refreshed at the end of three years.

The Distributed Approach

We assumed the distributed environment used 24 16-core HX5 blade servers to handle the 500 Linux and Windows applications. Since these environments require shared-nothing storage, the Storwize solution ends up requiring 126 enclosures and 1,285 raw TB of storage. All the hardware was purchased with the financing of the purchase price spread out over the five-year period. To meet the additional capacity demands year-over-year, new servers or storage arrays were purchased using the same methodology.

The Mainframe Solution

We configured a zEnterprise z196 model 501 to handle the database management, along with 13 IFLs and a zBX containing 14 16-core HX5 blade servers. The only application in the CP is the DB2 database management package. None of the distributed applications are rewritten to run on the CP. The Linux applications are relocated to IFLs, where there’s better memory management, allowing for greater utilization (up to 60 percent) and performance. We assume that 10 Linux applications can run on each IFL. Due to the improved management capabilities of a zBX, we assume a 10 to 15 percent performance improvement per HX5 on the zBX compared to a standard distributed environment.

In the zEnterprise environment, the data I/O requests start from the applications in the IFLs and zBX blades and are relayed to the DB2 application in the CP for handling. Only the DB2 application in the CP interfaces with the Storwize storage arrays. This environment initially requires 21 shared-storage Storwize enclosures and 214 raw TBs of storage.

At the end of the three-year lease, the zEnterprise model 501 is upgraded to a model 601 so it could handle the database workload through the next three-year period. As is common when upgrading a mainframe, IFLs are also upgraded. The cost to upgrade each IFL is $6,000 and is factored into the new lease. When the HX5 blade servers are upgraded at the end of the third year, the number of servers shrinks by two. We assumed that even though there are two fewer servers in use in year four, the licenses and associated software maintenance should be continued. This way, when it’s necessary to add more servers in the last year of the analysis, only the software for two servers needs to be factored in instead of four.

Using the previous scenario, we find that, as expected, the cost of the mainframe environment exceeds that of the distributed x86 servers by $9.4 million to $5.3 million on a Net Present Value (NPV) basis. However, the $4.1 million differential is more than recouped on the storage side. The zEnterprise storage costs come in at $3.8 million on an NPV basis while the distributed storage costs exceed $21.7 million. This is a net savings in excess of $13.8 million. Moreover, this savings is more than the cost of the entire zEnterprise ecosystem.

Analysis Considerations

The TCO analysis was done over a five-year period. On the leasing side, the original zEnterprise processors (CP, IFLs and HX5 blades) are returned after 36 months and replaced by the latest-generation servers. By swapping out the old hardware and moving to more powerful processors, the CP growth is contained and excess capacity is minimized. The IFLs growth is retarded and maxed out at 25 while the HX5 blades shrink initially upon replacement and then expand to a total of 22 blade servers. The Storwize arrays grow from the initial 128 TBs (214 raw TBs) to 314 TBs (523 raw TBs). However, the number of enclosures only grows from 21 to 28. This small expansion is the result of leasing the storage and replacing the units with more dense storage at the end of the three-year lease period.

The purchase model assumes that all servers are kept in service for a full five-year cycle and that, whenever added capacity is required, additional servers are bought. Thus, with the purchase model, the 24 servers slowly expand at a 20 percent rate annually until it reaches 48 servers by the end of the five-year cycle. The Storwize arrays expand from an initial 771 TBs (1,285 raw TBs) to 1.9 PBs (3.14 raw PBs) over the five-year period. The number of enclosures jumps from 126 to 216 in the same period, as none of the arrays or enclosures are swapped out.

On the software side in the purchase model, we assumed that payments for all software licenses were financed over the five-year period. However, in the leasing model, the costs of software licenses were spread out over the term of the lease. The leasing model selected was a Fair Market Value (FMV) lease obtained from IGF at a reasonable, but not most favorable, lease rate. The cost of capital and the purchase financing rate were estimated to be 6 percent.


We found it’s more than 100 percent more costly to distribute database serving among the distributed x86 servers than to consolidate the databases onto a common shared database platform using the zEnterprise as a database server. This cost savings is true on a current dollar basis and NPV basis.

The primary inhibitor to selection of the zEnterprise database engine approach is the fact the zEnterprise server alternative is more than twice as expensive as the x86 servers. Business and IT executives see the price tag differential—$1.14 million for the x86-based servers vs. $3.88 million for the zEnterprise servers over the five-year period—and conclude mainframes aren’t the way to go. However, the server costs pale when the database environment is factored into the equation. The cost for the distributed shared-nothing x86 storage systems comes in at $10.7 million while the mainframe storage system only costs $2 million over the five-year period. The $8.7 million savings in storage acquisition costs more than compensates for the $2.74 million in added zEnterprise acquisition expenses.

When all the TCO factors are examined, the purchased x86 solution runs almost $34.8 million, or on an NPV basis, just over $27 million. The leased zEnterprise solution comes in at more than 50 percent less—$16.5 million on a current dollar basis, or $13.2 million on an NPV basis.

The zEnterprise solution costs remained fairly flat over the five-year period, with most of the yearly expenses in the low $3 million range. There were two years when that didn’t occur—years three and four, where the expenses jumped to $4 million and then dropped to $2.7 million. The purchased x86 solution saw its total annual costs climb from $4.8 million in year one to $9.4 million at the end of the five years (see Figure 2).


The out-of-pocket charges to install the zEnterprise alternative is a wash compared to the installation costs of the x86 solution. However, there’s an $18.2 million savings that’s achieved by using the mainframe as a database server. Approximately one-third of that is hardware costs while another 27 percent savings comes from administrator costs.

In the purchased option, there was a requirement for additional software licenses and maintenance fees and growth in energy consumption. The total additional software expenditures in the purchase model exceeded $3.4 million, with most of that being software license and warranty fees. Similarly, power and cooling charges increased by more than $811,000 over the five years in the purchased model, or about 4 percent of the added expenditures (see Figures 3 and 4).

Other Considerations

There are several other advantages the zEnterprise platform offers that weren’t included in the cost analysis. Some of these are server-related while others are tied to the compressed storage footprint. Having just one copy of data reduces the risk of data integrity exposures caused by application or timing errors. This eliminates the need for syncing copies, which can consume between 25 and 45 percent of administrator time. Most companies today are concerned about the shrinking backup window; eliminating synchronization frees up time for backups. Companies often are hard pressed to get all their backups done as scheduled and are exposed, should a backup run fail to complete. There’s little time for a rerun. If a recovery is necessary, the most recent recovery point may not have been captured, potentially causing data integrity problems, lost revenues and customer dissatisfaction.

zEnterprise processors are architected for maximizing throughput and system utilization when consolidating multiple workloads on a server complex. Mainframes can consistently handle utilization levels of 80 to 100 percent without freezing or failing. Moreover, mainframes are recognized as the best platform for continuous and high availability, investment protection, performance, reliability, scalability and security. Because of its unique scale-up architecture, the cost per unit of work on a mainframe goes down as the workload increases; that isn’t the case with the scale-out architecture (see Figure 5). The cost/performance gains are due to the need for fewer administrators per unit of workload and higher levels of utilization. Mainframes can achieve higher utilization levels because of memory and processor sharing. Under the covers, there are hundreds of I/O processors to handle the data movements, freeing the central and specialty processors to focus on the application and task workloads.

This analysis didn’t examine the added costs of development systems. Here, too, the zEnterprise environment can share databases while each of the x86 test systems would have its own copies of the data. Moreover, users archive and back up the various databases and create snapshots. As shown in Figure 1, these database duplicates increase the rate of storage growth in the distributed environment over that of the mainframe solution. If these additional costs were added to the TCO, the zEnterprise advantage would improve even more.


We found that the zEnterprise reduced costs in all the TCO factors considered. zEnterprise hardware costs were 33 percent less than the x86 ecosystem costs while administrator costs were 28 percent lower. Warranty costs were 16 percent less and the cost of software dropped by 12 percent when the mainframe alternative was used. For much smaller SMBs or departmental systems, mainframes aren’t the answer, but for midsize to large enterprises, the economies of scale provided by mainframe solutions make a compelling case for organizations to re-examine their assumptions and consider the zEnterprise as a target environment.
Mainframe myths have led to higher data center costs and suboptimization. Organizations running hundreds of applications and multiterabytes of data should re-evaluate their architectural platform choices and evaluate whether or not a zEnterprise solution might provide them with a lower TCO. IT executives should insist on an evaluation that addresses the financial facts and ignores the religious platform wars. In today’s environment, IT must select and implement the best target platforms. The zEnterprise as a database server is a great choice.

No comments:

Post a Comment