Unix-to-Linux Migration: What Have We Learned?

Unix-to-Linux Migration: What Have We Learned?
Unix-to-Linux Migration:
What Have We Learned?
Your company is likely one of thousands running
Linux, which became a popular option in the past
two decades for edge computing (Web, firewall,
file/print and mail servers) because it was more
cost effective than Unix, or even Windows. Today,
almost all corporations have been involved in
Linux’s evolution from edge-computing solution
to a suitable replacement for Unix.
Until recently, x86 servers running Linux lacked the
performance; reliability, availability and serviceability
(RAS); and workload management of the high-end,
proprietary RISC/Unix platforms used to run missioncritical database applications. Those factors are no
longer barriers to the deployment of Linux to run
critical business applications. Today, enterprises are
migrating from Unix to Linux platforms running new,
multi-core x86 servers such as the HP DL980 that
overcome these limitations.
Many corporations choose to begin a Unix-to-Linux
migration for several reasons:
pThe cost (and economics) of x86 processors
is significantly lower (and better) than that
of RISC/Unix platforms.
pPerformance has increased to the degree
that Linux surpasses Unix in many standard
pIndependent software vendor (ISV) enthusiasm
has grown to the level that thousands of business
applications, including enterprise resource
planning and supply chain management tools,
are available on Linux.
pLinux is the leader in virtualization and cloud
computing — technologies that are changing
the way organizations function — which Unix
has virtually no chance of competing with,
except in isolated situations.
What We Have Learned
Linux dominates in Tier 1 and Tier 2 of three-tier
architectures. The biggest challenge left for the
platform is to remove any doubt that it can replace
Unix in hosting mission-critical database management applications that exist in the back-end third
tier of the architecture.
X86 servers need the vertical scale of RISC servers:
Initial x86 systems did not provide the vertical scale
of RISC/Unix servers. They had limited CPU core
counts, so the CPU and memory-intensive database
tier was left to the older Unix platforms, even though
these platforms were much more expensive than
Linux systems. In addition, when the CPU core count
in the x86 systems was low and virtualization was
deployed, there were unintended consequences such
as increased storage costs, complexity and inefficiency. Today, new x86 servers such as the DL980 have
vertical scaling capabilities that rival RISC servers.
Linux vendors are working to replicate the Unix model
for integrated capabilities: Unix is viewed as a highend, integrated stack of functionality, while Linux
has often been thought of as a platform in which
you have to put the pieces together yourself. For
example, Unix operating systems have built-in
disaster recovery and tight coupling with clustering.
When viewed as a replacement for Unix, Linux treated
these capabilities as auxiliary components. This
view of Linux is changing with the development of
new Intel x86-based Xeon hardware, which includes
many of the features of the RISC hardware used so
successfully to run Unix. Vendors such as Red Hat
are replicating the Unix model with add-ons tightly
integrated with Red Hat Enterprise Linux (RHEL) 6
for capabilities that are important for resiliency
and RAS.
Not all vendors, however, are following the lead of
Red Hat. VMware is saying that it has demonstrated
that virtualization is a separate market. Likewise,
Symantec is saying that it is proving that cluster
software is a separate market.
New Intel x86-based Xeon servers are competing with
big RISC servers running Unix: It’s becoming difficult
to differentiate RISC servers, such as the IBM Power
770 and Oracle Sun M9000, for most mission-critical
database applications, with x86-based servers
such as the HP DL980 using Intel Xeon 7500
series processors.
What’s compelling about the DL980 is that there
are application environments, such as databases,
where scaling is absolutely critical, and the DL980
can handle the scaling requirements. What’s more
beneficial is to be able to scale a database application vertically rather than horizontally. The DL980
gives you this capability.
The DL980 is an eight-processor machine, with each
processor available with 10 cores for a total of up
to 80. Much of the scalability and increased RAS
capability of the DL980 results from its utilization
of the HP Superdome node controller.
A good question is: Can RHEL 6 be married to the
DL980 as well as AIX has been married to Power
machines? The answer is yes. Red Hat is taking
advantage of the features that Intel has put into
its Xeon processors and that HP has put into the
DL980. Red Hat knows how to leverage the machine
check architecture feature set.
Another feature of the DL980 is the PREMA architecture, and what it does with dynamic traffic routing
and bottleneck detection. The PREMA architecture
provides smart CPU caching, increasing processor
scaling by reducing processor overhead and making
the DL980 competitive with RISC servers. It also
provides large cost efficiencies over RISC servers
running Unix.
We used to differentiate Unix with resource management because big Unix systems were going to run a
lot of different workloads. RHEL 6 has the resource
management capabilities that you can articulate
in Solaris, such as containers and scheduling. And
RHEL 6 supports virtualization with KVM integrated
with the kernel, giving it another way to do resource
management that is unavailable on most Unix systems.
Intel’s Xeon and Itanium processors share a lot
of the infrastructure in the way that they are linked
together. The I/O hub, the QuickPath interconnect
and the RAS features, along with the machine check
architecture feature set from Itanium, were incorporated by Intel into the Xeon processors. The Itanium
machine architecture provides a mechanism for
detecting and reporting system bus, ECC, parity,
cache and other hardware errors.
Oracle is also a big player in Unix-to-Linux migrations,
designing its database infrastructure (e.g., Oracle RAC)
so that it will run on only Oracle’s Linux platform.
Linux vendors are taking advantage of new x86 hardware features: Red Hat is integrating RHEL 6 with
the DL980 hardware, setting the stage for RHEL 6
to replace Unix as the platform of choice for missioncritical database applications. In the past, you could
differentiate, say, Solaris, from Linux by such factors
as performance, RAS and resource management.
Today, the Linux distributions are built on the 2.63
Linux kernel, which has good performance and
scalability. If you do a feature-by-feature comparison
of RHEL 6 with Solaris or HP-UX, you will find that
there is very little missing in RHEL 6 that is available
in the Unix counterparts.
Linux life-cycle support expectation is vastly improved:
Linux distributions from Red Hat for RHEL 6 are
offered with 10 years of production phase support,
followed by a three-year extended life phase. The
RHEL 6 life-cycle support is designed to reduce the
level of change within each major release and make
release availability and content more predictable.
The Oracle Linux life-cycle support also covers
a time frame of 10 years. IBM’s AIX has a 10-year
life-cycle support period, and HP-UX has 10 years
of life-cycle support that can be extended for an
additional three years.
Linux releases are not published as frequently as
they once were. For example, a new release announcement of RHEL occurs about every three years. Linux
releases are now much better coordinated with the
ISV community and with user organizations. This has
helped increase the number of applications available
on Linux into the thousands. New releases of ISV
applications will be on Linux and not on Unix.
Linux clustering can mask individual server highavailability issues: Many of the mission-critical
applications running on Unix are running on Unix
clusters. Clusters are used to protect against failure
in a system, placing availability requirements such
as five-nines for a mission-critical application
on the cluster versus on individual servers.
If you can meet your uptime service-level agreement
with Linux clustering, then the availability question
becomes harder to justify with Unix because Linux
clustering is just as good as Unix clustering. Competitive Linux cluster technology is available via Veritas
Cluster Server from Symantec and HP Serviceguard.
Organizations are migrating to Linux to consolidate
multiple Unix platforms: Server consolidation enables
users to consolidate multiple Unix servers onto a
single virtual host running Linux, leading to reduced
costs and often increased business agility. The first
step is to migrate a Unix server to a Linux server,
and then perform server consolidation by moving
the newly created Linux server workload to a virtual
server on a Linux-based virtual host server, running
KVM, Xen or VMware vSphere hypervisors.
Most of today’s popular virtualization technology
is x86-based and essentially unavailable on Unix
platforms. KVM is integrated with the Linux kernel,
and Xen is open source and can be used for free.
The DL980 is an excellent Linux-based virtual host
server because of its vertical scaling capabilities.
From the beginning, Linux had its challenges — until
they were met, users were reluctant to deploy the
OS beyond edge computing and perhaps consider
Unix-to-Linux migrations. These challenges included:
pChange from an open source, publish-often
life-cycle management paradigm to an enterprise
operating system that meets the needs of corporate
users and ISVs.
pImprove RAS.
pIncrease performance.
pGain ISV acceptance.
pSupport database applications.
During the past few years, we have seen the Linux
community close the gap with Unix with the advent
of x86-based servers such as the HP DL980 with
excellent vertical scaling capabilities and the marrying
of Linux distributions with these servers. The implementation of most of the features that made RISC/
Unix the platform of choice for mission-critical,
database-oriented applications, requiring high
reliability, availability and scalability, are now
available with selected x86/Linux servers.
Was this manual useful for you? yes no
Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Download PDF