zEnterprise vs. Intel Server Farms

How many Intel x86 servers do you need to match the performance of a zEnterprise and at what cost for a given workload? That is the central question every IT manager has to answer.

It is a question that deserves some thought and analysis. Yet often IT managers jump to their decision based on series of gut assumptions that on close analysis are wrong. And the resulting decision more often than not is for the Intel server although an honest assessment of the data in many instances should point the other way. DancingDinosaur has periodically looks at comparative assessments done by IBM. You can find a previous one, lessons from Eagle studies, here.

 The first assumption is that the Intel server is cheaper. But is it? IBM benchmarked a database workload on SQL Server running on Intel x86 and compared it to DB2 on z/OS.  To support 23,000 users, the Intel system required 128 database cores on four HP servers.  The hardware cost $0.34 million and the software cost $1.64 million for a 3-year TCA of $1.98 million. The DB2 system required just 5 cores at a hardware/software combined 3-year TCA of $1.4 million

What should have killed the Intel deal was the software cost, which has to be licensed based on the number of cores. Sure, the commodity hardware was cheap, but the cost of the database licensing drove up the Intel cost. Do IT managers wonder why they need so many Intel cores to support the same number of users they can support with far fewer z cores? Obviously many don’t.

Another area many IT managers overlook is I/O performance and its associated costs. This becomes particularly important as an organization deploys virtual machines.  Increasing the I/O demand on an Intel system uses more of the x86 core for I/O processing, effectively reducing the number of virtual machines that can be deployed per server and raising hardware costs.

The zEnterprise handles I/O differently. It provides 4-16 dedicated system assist processors for the offloading of I/O requests and an I/O subsystem bus speed of 8 GBps.

The z also does well with z/VM for Linux guest workloads. In this case IBM tested three OLTP database production workloads (4 server nodes per cluster), each supporting 6,000 trans/sec, Oracle Enterprise Edition, and Oracle Real Application Cluster (RAC) running on 12 HP DL580 servers (192 cores). This was compared to three Oracle RAC clusters of 4 nodes per cluster with each node as a Linux guest under z/VM . The zEC12 had 27 IFLs. Here the Oracle HP system cost $13.2 million, about twice as much as on the zEC12, $5.7 million. Again, the biggest cost savings came from the need for fewer Oracle licenses due to fewer cores.

The z also does beats Intel servers when running mixed high- and low- priority workloads on the same box. In one example, IBM compared high priority online banking transaction workloads with low priority discretionary workloads.  The workloads running across 3 Intel servers with 40 cores each (120 cores total) cost $13.7 million compared to z/VM on an zEC12 running 32 IFLs, which cost $5.77 million (58% less).

Another comparison demonstrates that core proliferation between Intel and the z is the killer. One large workload test required sixteen 32-way HP Superdome App. Production/Dev/ Test servers and eight 48-way HP Superdome DB Production/Dev/Test for a total of 896 cores. The 5-year TCA came to $180 million. The comparable workload running on a zEC12 41-way production/dev/test system used 41 general purpose processors (38,270 MIPS) with a 5-year TCA of $111 million.

When you look at the things a z can do to keep concurrent operations running that Intel cannot you’d hope non-mainframe IT managers might start to worry. For example, the z handles core sparing transparently; Intel must bring the server down.  The z handles microcode updates while running; Intel can update OS-level drivers but not firmware drivers. Similarly, the z handles memory and bus adapter replacements while running; Intel servers must be brought down to replace either.

Not sure what it will take for the current generation of IT managers to look beyond Intel. Maybe a new business class version of the zEC12 at a stunningly low price. You tell me.

BTW; are you planning to attend IBM Edge 2013 in Las Vegas, Jun 10-14? There will be much there to keep enterprise data center managers occupied.  Overall, IBM Edge 2013 will offer over 140 storage sessions, over 50 PureSystems sessions, more than 50 client case studies, and sessions on big data and analytics along with a full cloud track.  Look for me in the Social Media Lounge at the conference and in the sessions.  You can follow me on Twitter for conference updates@Writer1225.  I’ll be using hashtag #IBMEdge to post live Twitter comments from the conference.

Tags: , , , , , , , , , , , , ,

6 Responses to “zEnterprise vs. Intel Server Farms”

  1. Luis Fernando Lopez Gonzalez Says:

    Some months ago, knowing that software licenses could be more expensive than hardware would have been a surprise for me.

    Examples of TCA for the product you mentioned in the article are awesome and hopefully make some IT managers think twice what is the best option for his company.

  2. Infinity Systems Software Says:

    […] analyst Alan Radding published a positive blog post that highlights the technical and economic advantages of System z vs. Intel-based servers for […]

  3. Open to Open Systems Says:

    What about skills, dependency on legacy technologies that are dying, who’s going to be around to maintain and convert old z/OS applications in the future? I have no doubt that existing z/OS applications that are well established do a great job at a reasonable cost, but don’t kind yourself. No new systems or applications should be based on z/OS, it’s a step backwards. While I have any influence in the IT decisions that are made in my work place, the tide will be firmly against z/OS as a platform for next-generation applications.

    • Not Your Grandpa's Mainframe Says:

      I sure hope you don’t work for a bank, credit card company or anywhere my personal assets and information is stored. It is quite foolish to dismiss the mainframe as a antiquated computing system that isn’t capable of being the most efficient, affordable, secure, reliable, available, scalable and flexible computing system for “next-generation” applications. You would actually be surprised how many legacy clients are now moving applications to JAVA and deploying new workloads on z/OS and Linux Servers that run on the mainframe. There is a reason IBM is building and shipping more MIP capacity / mainframe processors worldwide than ever before. I’d challenge you to spend a little time with someone who truly understands the full capability of today’s mainframe and you will start to see what I mean.

  4. System z Wins IBM Platform Financials Race | DancingDinosaur Says:

    […] has been telegraphing the arrival in 2013 of a new business-class version of the zEC12 here and here, a z114 equivalent for the zEC12.  If it arrives next quarter, it should give a kick to z […]

  5. Sethuraman R Says:

    Thanks for sharing this post with us…nice work…ROC Software

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s


Follow

Get every new post delivered to your Inbox.

Join 655 other followers

%d bloggers like this: