Can virtualization stand the real world challenges of running resource-intensive applications such as SAP ERP in the enterprise level? Read on if you are part of a data center team or some IT department, part of enterprise architecture team, in charge of or want to help with defining your enterprise IT strategy, or just want to know a bit about what you can do with virtualization. Otherwise, please skip it – a newspaper and a cup of coffee would be more fun for you than reading about a POC on running SAP in virtualized environment.
Depends on who you talked to and at what time, you may have heard many different stories.
Familiar with something like below?
“Go ahead with you sandbox environment. Virtual machines work pretty well here. It’s cheaper, easier for management, and you can move it if your old hardware piss you off. At end of the day, if it really screws up, there are only a few developers you need to worry about. “
“But…, if you are talking about SAP production environment, big no until you have a safe place to hide from your frustrated end users, thousands of them, who may suffer from slow system performance syndromes”
A few years back, I won’t blame you for falling into such stories about virtualization. Think about what you can do with the leading player VMware and their flagship ESX server. If you still live in the ESX 3.0 time, no wonder you have doubt – what you can expect from a virtual machine that only supports 1 to 2 CPUs, up to 4GB memory, and up to 100 IOPS?
As the hardware getting more and more powerful in the last a few years, VMware has grown along the way. For example, ESX 3.5 had gained some muscle with its support of 4 vCPus, 64GB memory, and 100000 IOPS per VM. You can run some serious application on such a VM and I would not be surprised to see groups of SAP users use ESX 3.5 VMs for DEV, QA systems, or even some PROD systems.
The real challenge is to support PROD environment at large enterprise level where you are talking about thousands of users accessing mission critical applications and every minute of down time of a critical PROD system may implies millions of real $ loses.
Interesting enough, you are not the only one who wants to know the answer. We actually got a request from a major enterprise customer to find it out if the latest Intel processor and VMware platform available on the market can raise our expectation to an even higher level. A team from Intel, VMware and SAP Co-innovation Lab (COIL) took the challenge, set up an environment that closely emulated the customer’s real time environment, and tried it out.
I don’t claim myself to be any expert for virtualization or system performance measurement, but as part of COIL, I was lucky enough to be part of the team and worked closely with a group of real experts from Intel, VMWare, and SAP to tackle this challenge. I witnessed how the team worked day and night to test out the environment for a couple of months. BTW, talking about working hard, ask my colleague Shuuji Watanabe how he magically managed to not sleep or only sleep less than 4 hours a night for a few days when the testing was white hot. Igor Khurgin and Siva Gopal Modadugula can tell you how they felt like to work in middle nights to support Shuuji across the Pacific Ocean. Our friends Raghu Yeluri, Alan Bumgarner, Scott Allen from Intel, and Joachim Rahmfeld from VMware can tell you their fantastic passion for driving the best results from the project.
Personally I think the most interesting part of the exercise is that the environment we tested is designed to emulate the real customer environment, not something designed for benchmarking or for getting the best possible performance data. We even used an old version of SAP ERP (ECC 5.0) to emulate the real world scenario of running Sales and Distribution business processes with various load levels (hey, customer, an upgrade to your SAP app is due, EPR 6 rules nowƒº). Since the customer would like to find out what to expect from running ECC 5 with the latest and greatest from Intel and VMware, we brought into COIL a Nehalem white box (now it’s called officially Intel® Xeon® processor 5500 series) and VMware vSphere™ 4 to build the landscape.
The team tested various landscape configurations and put a lot of load to the landscape to find out how many concurrent users it can sustain and still give less than 1 second response time. It was a lot of work, but the team was excited about what they have found out. You know what? The team is definitely not shy about the results of the POC – they have actually published a whitepaper today on SDN to present the landscape and the performance measurement data.
The paper is downloadable from here: : http://www.sdn.sap.com/irj/scn/index?rid=/library/uuid/60458fab-e8f5-2c10-409e-d53eedcb7d54
I don’t want to draw any conclusion for you – will let you make your own judgment with the facts presented in the paper, including the nitty-gritty of the landscape configurations the team tested, and the corresponding performance data for each configuration.
Enjoy the reading.