Choosing between an x86 or UNIX system

Definitions of UNIX and x86 systems

X86 systems are derived from the 8086 chipset with the main processors being manufactured by INTEL and AMD. The chips are assembled into desktops, laptops and servers by hundreds of different vendors. In the enterprise server market space there are fewer manufacturers, with the following taking the biggest share of the market.

• IBM– x-series servers (Click to view products)
• HP– Proliant DL/ML and BL servers
• NEC– Express servers
• FUJITSU– Primergy servers
• Dell– PowerEdge
• SUN (Oracle)– Sun Fire and Netra

UNIX is a certification for a group of operating systems including AIX, Solaris and HP-UX that tend to originate from RISC based hardware. With UNIX the OS and Hardware tends to be developed by the same manufacturer.

The primary manufacturers of UNIX systems are:

• IBM with the POWER series currently available on POWER6 and POWER7 chips running AIX
• HP with the Integrity range of servers using the Itanium Chip , co developed by HP and INTEL running HP-UX
• Oracle-SUN systems with the SPARC chip running Solaris

SUN have confused the definitions by porting their Solaris operating system to run on x86 systems, however SUN still differentiate between x86 and RISC based SPARC systems by referring to the SPARC range as Enterprise systems.

UNIX systems’ position in the server market

If you consider the market based on size of system, the lower end is dominated by x86 systems and at the top end you will see mainframe systems. In between is a big mix of UNIX and x86 systems.

The improvement in performance and reliability of x86 systems has meant that there is now a large overlap between UNIX and x86 systems in the mid market. If you have a completely new system requirement without the challenges of migration from an existing environment you will have a bewildering number of options to choose from. If you are proposing to support between 500 and 1000 users you will be able to consider both x86 and UNIX systems, in scale out or scale up architectures, rack server or blade based systems.

If you have an established system that is due for a hardware refresh the decision is more difficult. If the system is business critical the risk in migrating will put a different perspective on the perceived advantage of a platform change. The cost of retraining existing supporting staff will also be part of your consideration.

Prioritise your system objectives

You need to be clear about the priority of your objectives for the system before you begin your selection. The priority of the system objectives will flow down from your business objectives and the role of role of the server in the business. Online transaction processing servers that are critical to order processing and invoicing need to be reliable and highly available. Data warehouse systems might be permitted a day’s outage but need the performance to run complex reports within a critical time window. The objectives you need to consider and prioritise include:

• Performance objectives – response time, batch run times, number of users supported.
• Reliability – how important is it that unplanned outages are minimised?
• Scalability – how big is the system likely to grow?
• Availability – does it need to be clustered, how long can I survive without it for maintenance?
• Total cost of ownership – what is the budget and is the hardware refresh cycle 3 or 5 years?
• OS preference for support – are there benefits in standardisation across the enterprise?

There are some situations that restrict the choice between UNIX and x86.

• Microsoft applications will not run on UNIX systems (with the exception of a small number of applications that will run on a Windows 2008 virtual machine under HP-UX).
• Software vendor support – some software vendors have a restricted list of supported hardware and OS.
• To keep the number of supportable platforms to a minimum some corporations have an OS strategy.

Reliability advantages of UNIX systems

UNIX systems’ Hardware and OS are designed by the same vendor. The objective is to provide the most reliable hardware and provide the most stable, scalable OS possible. The Hardware in UNIX systems offers greater levels of protection against component failure. An example of these features is IBM and HP UNIX systems’ ability to survive a processor failure without halting the OS. The vendors of UNIX systems spend more time testing 3rd party adapters with the hardware and operating systems to ensure the maximum level of reliability. x86 systems tend to rely on the 3rd party vendors to do the testing for different hardware and operating systems. This does mean that the components supplied for UNIX systems tend to be more expensive than x86 systems.

Improved adapter and peripheral support of x86 systems

Because the components for x86 systems are developed and tested by the 3rd party vendors the list of available adapters and peripherals is vast in comparison to UNIX hardware, often with less expensive alternatives. x86 systems also tend to be quicker to adopt the newer speeds and latest technologies before UNIX systems.

x86 and UNIX systems performance comparison

Finding benchmarks to compare performance of x86 and UNIX systems can be a challenge. The published benchmarks tend to quickly become out of date and not all vendors choose the same benchmarks. Each vendor will also not release benchmark results for all of the servers in the range; often just for one of the larger flagship models. It is important to pick a benchmark that reflects the requirements of your application as they are each designed with a specific applications purpose in mind. Some of the benchmarks available are:

Benchmarks

SPECjbb2005 is SPEC's benchmark for evaluating the performance of server side Java.

SPECjAppServer200 is designed to measure the performance of J2EE 1.3 application servers.

SPECjEnterprise2010™ benchmark is a full system benchmark which allows performance measurement and characterization of Java EE 5.0 servers and supporting infrastructure such as JVM, Database, CPU, disk and servers.

SPEC CPU2006 measures compute-intensive performance across a range of hardware using workloads developed from real user applications.

TPC-C is an on-line transaction processing (OLTP) benchmark. Throughput is measured as the average number of transactions processed per minute during twenty minute run. IBM tends to use DB2 on their servers to produce these results making it difficult to compare with the other results based on Oracle.

rPerf is IBM’s own comparative performance report for POWER systems. The rperf document also contains some of the industry standard benchmarks listed above.

SAP has its own sizing benchmark called SAP Standard Application Benchmark. For more information on the SAP Standard Application Benchmark and SAP sizing see SAP Sizing and The SAPS Benchmark.

Scaling up UNIX and x86 systems

UNIX systems scaling to 20-30 cores running large business critical systems have been common for several years. x86 systems scaling to this size have not been used in production for as long. One of the restrictions for the x86 design has been the limit of memory that could be addressed. The development of 64bit x86 architecture and OS has removed these limitations and much larger systems are now available.

The SAP benchmark results now list a Fujitsu x86 system running Windows Server 2008 R2 Datacenter Edition, with 8 INTEL Intel Xeon Processor X7560 processors totalling 64 cores with 512Gb of RAM. It is rated at 87,550 SAPS with 16,000 benchmark users.

For comparison the IBM POWER7 750 with 32 cores and 256Gb of RAM is rated at 85,220 SAPS at 15600 benchmark users.

So for the mid market the technical ability to scale up is available in both x86 and UNIX. The decision on preferred choice at this scale is likely to involve how much trust you have in the reliability of the operating system and hardware, which will depend on your personal experience.

If you are at the top end of the mid market you may be able to take advantage of the grow on demand features that most of the UNIX vendors offer (CuOD, TCOD and UCOD for IBM, TCAP, GCAP and PPY from HP), allowing you to have spare capacity on the floor that can be switched on for a fee when you need it, without interruption to service. With these features you can be more aggressive with your sizing; keeping cost to a minimum safe in the knowledge that there is more capacity to hand should you need it.

Scale out Architectures

One of the reasons scale up architectures have not been as common in the x86 market is because of the introduction of scale out technologies enabling you to implement multiple lower cost x86 systems. Web and Application tiers are well suited to scale out architectures to service a high number of parallel operations from across the network. These layers in the application stack tend not to hold data directly and so do not have the complexity of requiring communication between the nodes in a cluster and often sit behind a relatively simple load balancer distributing user requests. Application tiers running JAVA often demand high levels of memory which is likely to work out less expensive on x86 based systems.

Oracle’s scale out database architecture “RAC” has become a popular choice in the x86 market. Using multiple low cost x86 servers to get the performance and resilience you need is an attractive option, but there are some important costs that need to be considered alongside the purchase price of the servers themselves. Each node requires shared access to central storage so the SAN could be more expensive and more complicated. A high performance network (interconnect) is required to allow the nodes to communicate between each other to ensure the integrity of the data, provide cluster heartbeat signals and enable parallel execution of queries. These infrastructure elements will be critical to the overall performance of the system and will add to the complexity of the solution.

The performance per core will reduce each time you add an additional node to the cluster so you may end up needing significantly more cores to get the performance you need; this can have a significant impact on your overall TCO if you need the Enterprise edition. See Calculating the TCO of Oracle licences for infrastructure choice for more details.

Virtualisation features and limitations

The virtualisation technologies on both x86 and UNIX systems are keeping pace in terms of features. The ability to migrate running systems between hardware is now available to POWER, Integrity and VMware x86 systems. One of the differentiators currently between x86 and UNIX systems is the maximum size of the individual virtual machine. With UNIX the only limitation is the size of the hardware; with VMware and HyperV there is a restriction on the number of cores you can run within your virtual machine. Another important factor for Oracle users is that Oracle licensing rules do not recognise VMware as a means to divide a system for licenced use. If you have a 16 core system running VMware and one virtual machine of 4 cores running Oracle you will need to licence the full 16 cores. Most of the UNIX based virtualisation technologies do not suffer this restriction making them a better choice for consolidating the entire software stack with each layer operating as a virtual machine/ partition. The cost of the virtualisation layer as well as the guest OS images need to be considered in your TCO calculation.

View our full catalogue of IBM System X appliances: Click here

To speak to one of our Specialist Consultants, call us on 0115 951 9666, or add your details above and we'll get right back to you.