MAINFRAME > TRENDS > LINUX

Businesses See an Advantage to Using Linux on z Systems Over Scale-Out Servers

Linux on z Systems
Illustration by Wayne Mills
 

The IBM z Systems* mainframe is well established today as the invaluable system of record (SoR) for mission-critical business data and applications. It has evolved over the past several decades to be the IT centerpiece of many of the largest banks, retailers, financial firms and healthcare companies in the world.

Millions of other servers are also in use. Server sprawl—driven by huge numbers of inexpensive, scale-out servers—has increased at a phenomenal rate. Since the introduction of virtualization in the mid-2000s, the number of VMs supported on those servers has also grown, with IDC reporting that x86 servers are nearing VM saturation. Although today’s most critical, complex and compute-intensive workloads are candidates for virtualization, these workloads also have ever more stringent requirements for performance and service delivery.

For instance, analytics workloads support large databases yet must also meet demand for real-time response. Large clusters of scale-out servers generally don’t have the sustained performance capacity or large-scale memory and cache requirements to adequately drive these workloads—database sharding (partitioning) has become popular as a way around this issue—and the vast network connectivity requirements introduce even more latency. Mobile workloads not only put unexpected and unpredictable performance strains on IT servers, but they also demand exceptionally fast response times. As businesses move beyond simple mobile connectivity and into the API economy, co-location of systems of engagement (SoE) services with SoR assets becomes critical to maintain immediate response.

A better solution is Linux* running on z Systems. IBM z* provides exceptional performance and service delivery to efficiently drive large workloads as well as consolidate large numbers of smaller workloads. In addition, when running Linux on IBM z co-located with z/OS*, hybrid workloads that span the SoE/SoR boundary run with highest efficiency, and often with lower cost.

Performance and Service Delivery

IBM z Systems servers are far more powerful than typical scale-out models. Mainstream Intel* cores support clock speeds of roughly 2.3 to 3.5 GHz, while IBM z13* cores deliver 5.0 GHz—the fastest commercially available—and IBM z13s* cores run at 4.3 GHz. The z13 tops out at 10 TB of memory, with the z13s capable of 4 TB. Each IBM z core has over 4.3x more on-chip cache per core than an Intel core. IBM z also has shared L4 cache, which doesn’t exist on Intel chips.

The impact of clock speed, memory and cache differences means that data-driven, cache-intensive, compute-intensive workloads run faster on Linux on IBM z than on x86 servers.

Database workloads are by nature very cache-intensive. IBM engineers ran a pure memory database access test on an IBM z13 with six cores running Linux. The same test produced virtually the same results on an x86-based server but required 80 cores. Because the test ran with so many fewer cores on the z13, the database software was calculated to be 6.7x more expensive on the x86 server than on Linux on z, according to IBM.

Emily Farmer is a senior research consultant in the IBM Software Group Competitive Project Office.


comments powered by Disqus

Advertisement

Advertisement

2019 Solutions Edition

A Comprehensive Online Buyer's Guide to Solutions, Services and Education.

10 Years With Tux

10 Years With Tux

3,000 and Growing

IBM encourages easy ISV application porting to Linux on System z

Successes Mount for Linux on System z

Usage patterns evolve, increasing value to customers

IBM Systems Magazine Subscribe Box Read Now Link Subscribe Now Link iPad App Google Play Store
Mainframe News Sign Up Today! Past News Letters