Mainframe Migration - 4
of data is moved between the disk and CPU. As we look at using Micro Focus Lift and Shift process for migrating applications off the mainframe to Windows-based environment the ratio logically continues to hold.
We are now seeing the emergence of “enterprise-class” Windows and Intel systems that are well balanced. These systems are similar to mainframes in terms of high speed network attachments, fiber optic interconnects to disk, large disk buffer caches (battery backed for transaction durability requirements), dual power supplies and bus architectures for reliability, tiered client networks and end-user stations relieving the host of low-level handling. It is this class of machines on which we focus our benchmarking efforts. These systems are very different from “personal” desktop or laptop computers where a 1000:1 ratio of CPU to I/O is not uncommon, instead of the desired 1:1 for business online transaction processing systems.
The 64-bit Intel and Windows architectures show promise to run even larger images and workloads. However, we focus on Intel XEON-based 32-bit systems. Most COBOL applications moving from the mainframe host still employ 32-bit (actually, 31-bit) designs. The ability of the operating system to manage multiple 32-bit environments and allocate memory resources that exceed the 32-bit virtual memory of any one process is important to scalability. Enterprise-class Intel hardware architectures, the Windows operating system, and Micro Focus Enterprise Server with Mainframe Transaction Option can exploit physical memory many times that of the 32-bit virtual address space and this is brought out in these scalability tests.
The enterprise-class environment used to verify scalability of the Lift and Shift environment was a Unisys ES7000 Orion. The architecture of this system allowed partitions to be set up consisting of various memory and processor configurations, just as IBM logical partitions (LPARs) or Amdahl multiple domain facility (MDF) are used to fence off workloads from one another (see following table). The mainframe migration scalability test environment consisted of partitions of 4, 8, 12 and 16 processor configurations, all but the first on the Unisys ES7000. The smaller 2-CPU partition test was conducted on a different architecture to compare smaller server machines. Although capable of 32-CPU configurations as a single system Windows image, the MIPS capacity of these Unisys images were too large to be driven by our test workload generation machine environment. The Unisys “system under test” partition required a 3x horsepower to simulate the end-user workload and gather response time measurements.
Unisys ES7000 Orion 540
Intel XEON 3.0 GHz
Switched Gigabit/s Ethernet Hub
Hitachi 9980V Storage Device
2 Gigabit/s Fiber
Note: The difference in the results here from those in the earlier paper are due to an update of the Unisys system (3.0 GHz XEON processors vs. the earlier 2.8 GHz) and a reconfiguration of the disk storage subsystem.
Transaction Processing Benchmarks
The transaction processing and database community utilize a series of benchmarks by the Transaction Processing Performance Council (TPC) to measure relative
WinHEC 2005 Version - April 20, 2005