| |
| Datatrend Newsletter: Q1 2010 |
The CEO's Corner
Okay. Another year is upon us. So, what are we going to do about it?
What are we going to do in 2010? Let me propose something. Let's elect
to influence this year in a positive way. It seems like a good idea to
me for us to have a great year.
[ Read the full article ]
Virtual SMP for High-Performance Computing - Turning Virtualization Inside Out... Literally!
High performance and computational-intensive applications have
progressively demanded faster and larger systems to perform the work.
SMP and MPP designs evolved over the last few decades to address the
expanding workloads.
[ Read the full article ]
Data Center Cabling and Design Best Practices
Datatrend has been in the structured cabling business for more than 20
years, and we have seen the best and the worst in cabling practices.
Our experience has shown us that even a small network infrastructure
misstep can lead to big costs and headaches down the road. Luckily,
with a little planning and knowledge, you can avoid major network
installation mistakes and ensure that your network infrastructure will
flawlessly perform its critical behind-the-scenes job for many years to
come.
Have you ever looked at an "unstructured" cabling implementation under
a data center floor? This spaghetti mess of wires raises some serious
questions:
[ Read the full article ]
Industry Standard (x86) Systems - a World of Innovation
On March 29, 2009, Intel introduced the Xeon 5500 microprocessor, a
radical departure from previous designs setting a new direction in the
architecture of the highest volume computer platform in history. In
fact, this CPU is such a huge leap forward in x86 server performance,
that it has enjoyed one of the most rapid product transitions ever,
perhaps because it provided the greatest generational increase in
performance the Xeon processor has ever experienced.
[ Read the full article ]
Contributed Article: App Management is the New War for Virtualization in 2010
In
an age of virtualization, cloud computing and newer application
architectures such as Web 2.0 and distributed service-oriented
architectures (SOA), the interdependencies inherent in today's
multi-tier applications have created innumerable points of failure that
will continue to impact IT management throughout 2010.
[ Read the full article ]
Tech Tip: Installing and Using Xen - Xen Virtualization Tutorial
Want to learn more about Xen? Click the link below to take advantage of this month's Tech Tip pearl of wisdom:
[ Read the full article ]
|
|
Featured Webinar
|
Virtual SMP for High-Performance Computing - Turning Virtualization Inside Out
Thursday, February 11, 2010
1 PM CST
Click here for more info
|
Contact Us
|
For more information contact Datatrend directly.
[ contact us ]
|
|
The CEO's Corner
Okay. Another year is upon us. So, what are we going to do about it?
What are we going to do in 2010?
Let me propose something. Let's elect to influence this year in a
positive way. It seems like a good idea to me for us to have a great
year.
To get on the right track immediately, let me ask how Datatrend
Technologies is doing in your eyes. Drop me or your primary contact a
note. We endeavor to positively exceed expectations, but we need your
feedback to be sure.
The pursuit of excellence starts with communication, collaboration, and
a partnership resolve to formulate the best approach to determining the
right solution. Certainly this takes some effort on the part of the
customer and the strategic alliance partner (vendor).
The more Datatrend understands your key business
initiatives/objectives, and your preferences & corporate standards,
the better we can serve you.
Your Datatrend team wants to invest time, effort, and resources in
meeting with you, as soon as convenient, in determining your 2010 focus
priorities and with respect to anything else you might want to share
with us.
In return, we will respond to that initial collaboration session with
appropriate follow-on activities. This might include
infrastructure/architectural whiteboarding sessions, granular deep-dive
technology briefings, trials/evaluations, test validation/proof of
concept pilots, development of IT solution benefits analysis or
assistance in developing ROI business cases related to the given IT
project undertaking (optimization, migration, network services, etc.).
Partnership results are impacted by level of talent, experience, quality of collaboration and effort.
Let's take hold of 2010 and dominate it by preparing for it and
outthinking it. Daring to be great is the suggestion. You have many
choices with whom to partner. Compromises might best be dealt with
during contract negotiations, but neither in work performance nor
results.
Datatrend would feel honored to be allowed to apply great effort in
collaborating with you and in responding thereafter with solutions that
you judge that hit the mark. Let's make 2010 a year we master in all
respects.
Respectfully,
Mark Waldrep
CEO Datatrend Technologies
mark.waldrep@datatrend.com
|
Virtual SMP for High-Performance Computing - Turning Virtualization Inside Out... Literally!
High performance and computational-intensive applications have
progressively demanded faster and larger systems to perform the work.
SMP and MPP designs evolved over the last few decades to address the
expanding workloads:
MPP - distributed memory systems.
Massively Parallel Processing systems require special programming
techniques involved with message passing between the application
fragments running on each processor.
SMP - shared memory systems.
Symmetric
Multi-Processing can refer to any shared-memory system, even ones that
use other memory architectures, as long as all system processors can
access the entire memory address space.
SMP systems became very
popular compared to MPP systems due to the ease of deployment,
management and programming. However, these systems were usually
expensive as they used custom processors, chipsets and ASICs to create
high-speed backplanes as well as sometimes using custom Operating
Systems.
Intel and AMD processors became much faster and offered a viable
alternative for proprietary server solutions especially with the
adoption of loosely coupled compute clusters. These "commodity"
servers, interconnected with standard high speed interconnects to form
clusters, are still similar in nature to MPP systems and require a more
difficult programming model to allow the application to span across
multiple systems. MPP systems deliver more performance than traditional
SMPs, as the distributed architecture provides bottleneck-free
performance for shared-nothing, or almost shared-nothing applications.
The need to implement and support multiple OS instances and
interconnect fabric technologies make these clusters more difficult to
deploy and manage, however.
Ideally, high-performance customers want a way to marry the great
price/performance of volume x86 system technology with the simplified
implementation and management of an SMP without the cost overhead of a
traditional large SMP system.
Enter high-end virtualization for aggregation from ScaleMP. ScaleMP's
versatile SMP (vSMPtm) architecture eliminates the distributed nature
of the cluster and enables unified compute resource optimization of an
SMP for capability, manageability and flexibility. The single virtual
SMP (vSMP) system is an x86 supercomputer today supporting 128 cores
and 4TB of shared memory running a single instance of a Linux operating
system. ScaleMP's vSMP Foundation™ software creates a single virtual
system by aggregating multiple blade or discrete servers,
interconnected via an Infiniband fabric and is ideally suited for high
performance computing applications for financial services, life
sciences, engineering and educational institutions.
The ScaleMP solution is available in three flavors: vSMP Foundation for
SMP, vSMP Foundation for Cluster, and vSMP Foundation for Cloud,
offering a flexible variety of scale and sophistication depending on
customer requirements. Each offers significant price/performance
advantages, power consumption savings, and higher density over
traditional and proprietary SMP systems. Finally, high performance SMPs
are affordable again.
How Does It Work?
Traditional SMP systems run a single operating system (OS) that
interacts with the system using a well-defined hardware interface such
as Intel's Multi-Processor Specification (MP Spec). For a traditional
SMP system, this interface is implemented in a silicon chipset. In
addition to the hardware interface, an SMP system consists of CPUs,
memory and I/O subsystems all connected together with an interconnect
such as Intel's FSB (Front Side Bus) or newer QPI, AMD's HT
(Hyper-Transport), SUN's CrossBar, SGI's NUMALINK, and IBM's eXA. The
system interconnect is where today's SMP systems differ the most from
one another.
ScaleMP Versatile SMP™
The vSMP architecture utilizes commercial-off-the-shelf (COTS)
components and does not require any custom parts. Its key value is the
utilization of software to provide the chipset services that are
otherwise required to create traditional multi-processor systems. vSMP
Foundation provides cache coherency, shared I/O, and the system
interfaces (BIOS, ACPI) that are required by the OS. The vSMP
architecture is implemented in a completely transparent manner; no
additional device drivers are required and no modifications to the OS
or the applications are necessary.
vSMP Foundation requires
Multiple x86 systems (rack servers or blades)
InfiniBand HCAs & switch to interconnect the systems or blades
vSMP Foundation virtualization SW - comes as flash memory inserted in each aggregated server or blade
vSMP Foundation provides
One single system: vSMP Foundation aggregates all the resources of the
multiple physical systems, initializes the interconnect fabric, and
creates the required BIOS and ACPI environment to provide the OS a
coherent image of a single virtual system. vSMP Foundation then uses a
software-interception engine in the form of a Virtual Machine Monitor
(VMM) to provide a uniform execution environment with no large SMP
premium price and no cluster management complexity.
Coherent Memory:
vSMP Foundation maintains cache coherency between the individual boards
using multiple advanced coherency algorithms that operate concurrently
on a per-block basis, based on real-time memory activity access
patterns. vSMP Foundation leverages board local-memory together with
best-of-breed caching algorithms to offset the effect of interconnect
latencies. This results in simple, high performance with simple
programming.
Shared I/O: vSMP Foundation
aggregates I/O resources across all boards into a unified PCI hierarchy
and presents them as a common pool of I/O resources to the OS and
applications, reducing storage costs and complexity.
Datatrend and ScaleMP
Datatrend is the IBM global solution provider for vSMP systems. Our
technical experience and integration capabilities provide cost
effective, reliable virtual SMP systems utilizing the latest Intel Xeon
processors and IBM's innovative BladeCenter and iDataplex platforms. We
can also design and implement to unique customer requirements for vSMP,
high performance storage and networks.
Datatrend provides consulting, design, integration installation and
support for vSMP solutions anywhere in the world. To see if a turnkey,
high performance, low cost virtual SMP solution is right for you, give
us a call or visit us at www.datatrend.com/sharedmemory
Click here to discover more about our virtualization services.
|
Data Center Cabling and Design Best Practices
Datatrend has been in the structured cabling business for more than 20
years, and we have seen the best and the worst in cabling practices.
Our experience has shown us that even a small network infrastructure
misstep can lead to big costs and headaches down the road. Luckily,
with a little planning and knowledge, you can avoid major network
installation mistakes and ensure that your network infrastructure will
flawlessly perform its critical behind-the-scenes job for many years to
come.
Have you ever looked at an "unstructured" cabling implementation under
a data center floor? This spaghetti mess of wires raises some serious
questions:
- When moving equipment, which wires do I unplug?
- How do I avoid disturbing other wires when I am working on wiring modifications?
- How much time and effort do I spend deciphering the connections every time I make a change?
Virtualizing your data center presents a great opportunity to migrate
to a structured cable environment, eliminating older technology devices
and their cables from the plant. If you are building a new data center,
you can get it right from the start and give your data center team the
time to move devices with confidence, knowing exactly how connections
flow and where they are attached. Using a wiring blueprint will save
countless hours of tracking wires and dramatically lessen the risk of
accidentally unplugging production devices.
Datatrend's 23-year practice has helped us develop a proven six-step
approach to help you lower cost and reduce risk with your cabling
projects.
Step 1 - Requirements Gathering:
Datatrend begins cabling engagements with a detailed "as is" assessment
of the environment. This includes site surveys and documentation of
your cabling conventions and historical practices, and development of a
detailed understanding of what you want to accomplish. This identifies
areas of risk and details the pros and cons of various alternative
plans, so you can choose the best approach to achieve your goals. This
is also the time to develop a budget that will reflect realistic,
actual implementation costs. Adjustments in approach can be made to
better accommodate budgeting constraints.
Step 2 - Implementation planning:
In the planning process, we look at the elements of the data center
environment and begin to build a cabling design plan. For new data
centers, planning cabling pathways from scratch ensures the most
efficient layout of cable paths. When working with an existing data
center, planning complexity increases. Cable trays may be at capacity,
production wiring is already place, and many possible areas of risk
need to be addressed in planning.
The goal of the discovery
process is to identify every server, switch or storage device and
determine the best pathways to connect all this equipment using as
little space as possible. This includes not only identifying the
equipment but also determining the connectivity requirements. Where to
implement optical cabling, for example, versus copper cabling? What are
the bandwidth requirements? For an existing environment, what is
currently used for connectivity? This may be more than adequate to
handle any new additions or you may need to expand the system. Are we
dealing with a single-purpose data center or one that hosts multiple
companies?
What about the physical space where the cabling will run? Are there
drop ceilings or are the cables running through a raised floor? How
many connections can be placed into a single rack and how many racks
are available? What other systems might affect your cabling choices?
This includes construction issues, heating and cooling systems,
electrical systems - even how many IT staff members need to interface
with the system. All these issues need to be considered in your final
design and implementation plan.
A network infrastructure can last up to 20 years, so it is important to
consider future needs in your project plans. Do you want to go
wireless? Will your network speed requirements increase? Where will you
need network connectivity? Spending a little more in the short-term can
prevent costly re-dos in the future. Of course, safety and code
requirements must always be part of the plan.
Step 3: Determine a Project Time Schedule:
Some technology components require a long lead time for delivery. Given
these constraints, components and/or plans may need to be altered to
accommodate the schedule.
Step 4: Finalize the Budget:
Understanding cost estimates throughout the planning process will avoid
over-designing what you can't afford. Cost estimates should include
physical devices as well as the implementation resources required for
the project.
Step 5: Create the Design Plan:
Once all design points and issues have been identified and resolved, the final design blueprint should include:
- Comprehensive floor plans and prints
- Identification of the main distribution areas (MDA)
- Identification of zone distribution areas (ZDA)
- Identification of equipment distribution areas (EDA)
All these elements should be well planned and reflect the correct
safety and code requirements for your state, including national, state
and local codes, as well as industry and manufacturer standards (BICSI,
NEC, NESC, TIA, EIA, IEC, ANSI and CSA).
Step 6: Installation:
This is the time to ensure that all implementation details are
identified, tasks assigned, dependencies understood, and resources
assigned with specific dates for each piece of work.
Designing
and implementing a data center cable system is often a task best left
to experts. Hiring a consultant such as Datatrend, who will work with
you to develop an effective plan and assign a project manager to
oversee every detail, can often be the most cost effective way to
guarantee that your cabling plan will best serve your needs for years
to come.
Datatrend specializes in full-service design, installation and
maintenance of network projects across a wide range of industries. We
have a solid track record building virtually every type of network,
including voice, data, and both fiber and wireless environments. If you
need assistance with your cable design, give us a call. Our skilled,
experienced consultants and engineers would be happy to work with your
project team to develop an optimal and cost-effective design,
customized to your needs and consistent with best practices.
To learn more about Datatrend's Network Services or to request a quote for your next project click here.
|
Industry Standard (x86) Systems - a World of Innovation
On March 29, 2009, Intel introduced the Xeon 5500 microprocessor, a
radical departure from previous designs setting a new direction in the
architecture of the highest volume computer platform in history. In
fact, this CPU is such a huge leap forward in x86 server performance,
that it has enjoyed one of the most rapid product transitions ever,
perhaps because it provided the greatest generational increase in
performance the Xeon processor has ever experienced.
But that was just the beginning!
The Intel Xeon 5500 (code named Nehalem EP) is a significant new core
design for Intel in their "tick-tock" rapid-fire development strategy,
this being the "tock" or architectural enhancement phase as opposed to
the "tick" or die shrink phase of an existing enhancement.
http://www.intel.com/technology/tick-tock/index.htm
With Nehalem came a new processor core and entirely new memory
subsystem architecture based on the direct attach Quick Path
Interconnect. These changes permitted radically more performance from
the same power footprint. But, in normal Intel fashion, this first
iteration of the new architecture only applied to two socket servers
and blades.
What's next for Intel and IBM? Two significant enhancements are on tap.
The lesser of these is the next "tick" for the dual processor version,
code-named Westmere.
Westmere takes the Xeon 5500 Nehalem EP core and drops from the current
45nm fabrication technology to 32nm, giving Intel the opportunity to do
more in the same physical (electrical and thermal) envelope. With six
cores and larger cache sizes, Westmere will boost heavily threaded
application performance and virtualization.
But this is the lesser of the two enhancements coming up. The
long-awaited refresh of the Xeon 7000 series is nearly here, and it's
significant.
Intel makes a variety of processors, but processors for servers can
roughly be categorized by socket capability: single socket Xeon 3000
sequence, dual socket Xeon 5000 sequence and greater-than-dual socket
Xeon 7000 sequence. There is still the Itanium processor, but market
demand for IA64 predicts that is does not have long term viability.
The Xeon 7000 is the flagship of Intel's processor line and really has
the greatest performance per core and the greatest scalability. This is
the processor family that takes the "pizza box" server and allows the
development of real enterprise-scale systems and innovation to get
interesting: especially what IBM has done in this space over the last
10 years..
The Intel Xeon 7000 CPU family always gets updated after the 5000s
because it is a more sophisticated system supporting significantly
greater resources. Traditionally Intel has provided the same "chipset"
(silicon infrastructure to connect CPUs, memory and IO) for dual-socket
and quad-socket systems that all manufacturers use. Building chipsets
is something of an art and requires deep systems architecture
engineering expertise, especially as the system design gets larger. In
2001, IBM partnered with Intel and introduced its own version of a
"quad-socket" chipset that could scale much larger. Since 1998 IBM has
been engineering this scalable Xeon chipset, refining it with large
systems know-how garnered over decades of systems design experience.
Today eX4, IBM's 4th generation of this chipset is the best-selling
quad socket architecture and dominates larger four-socket system
volumes in the market. Price, performance, scale, flexibility,
reliability and simplicity have characterized IBM's Enterprise
X-Architecture chipset program, and they aren't stopping just because
they're on top.
The new Nehalem EX, a multi-socket version of Nehalem, was previewed
back in May 2009, and we anticipate an eight-core CPU with much more
cache and even greater memory bandwidth and scalability than Nehalem
EP. Customers can expect a huge performance gain in these larger
systems, which will be optimal for database, virtualization and
large-application workloads. Add in IBM's decade of proven Xeon 7000
innovation experience, and we expect something truly exciting!
If IBM was able to set more than 100 #1 benchmarks with eX1 through
eX4, including a whopping 1.2M tpmC world record for x86 (using just 8
CPUs), what can we expect for eX5?
Gone are the days when HP-UX on Itanium or MIPS, Solaris on SPARC, or
any other "enterprise" proprietary system can claim scale as their
calling card. When x86 systems from IBM, namely the x3950M2, started
surpassing 1M tpmC, enterprise class and x86 became synonymous.
The new eX5 systems from IBM promise unparalleled scalability and
performance while introducing even greater levels of flexibility than
previously thought possible. The innovation IBM is applying to Intel's
Quick Path Interconnect, plus combining it with their proven scalable
port technology, will provide a building block system capable of
handling even the largest SMP and virtualization workloads. For the
first time, IBM is applying this technology to both modular rack
systems and BladeCenter.
As organizations look for new ways to save cost, enhance service levels
and increase IT responsiveness, the ability to virtualize larger
workloads without the headaches resulting from a huge number of servers
gives eX5 a unique value to the enterprise that is serious about IT
optimization and cost effectiveness. If you want to get really
creative, think of combining Nehalem EX, IBM eX5 and virtual SMP to
create a wide variety of enterprise and HPC platforms from a few
standardized building block technologies, exploiting the greatest range
of virtualization capabilities.
As Westmere, Nehalem EX, and IBM's eX5 systems come to market in early
2010, Datatrend can help you get a deeper understanding of how these
technologies can benefit your technology infrastructure. Since
Datatrend is up to speed on what is coming out through 2010, contact us
today and we can guide you on how to be prepared to receive the
benefits of these technologies as early as possible to maximize your
2010 return on investment.
If you want to learn more about these exciting new developments, please contact a Datatrend representative now: http://www.datatrend.com/contactus-form.html
|
Contributed Article: App Management is the New War for Virtualization in 2010
|
Contributed article by Vic Nyman, co-founder & COO, BlueStripe Software
In an age of virtualization, cloud computing and newer application
architectures such as Web 2.0 and distributed service-oriented
architectures (SOA), the interdependencies inherent in today's
multi-tier applications have created innumerable points of failure that
will continue to impact IT management throughout 2010.
Some virtualization management solutions interpret application
performance by measuring response times between virtual machines (VMs)
and clients. Others deliver end-user response times. Neither delivers
the insight needed by IT Operations and application support teams into
the physical realm and how that relates to the virtual infrastructure
supporting applications. Many organizations struggle to quickly and
conclusively determine if poorly performing applications are based on
issues related to the network, storage hardware, VMs, application
instances or some combination of those.
This cross-platform, cross-environment visibility is emerging as a
major problem for enterprises as their applications are supported by
existing physical servers and connect to a variety of operating
systems, applications and platforms. To further complicate the issue,
many IT organizations desire to move larger, more complicated and
business-critical applications onto a virtual infrastructure. When
unintended performance issues arise, isolating where and why the
problems are occurring becomes even more challenging.
The focus for measuring application performance on VMs needs to shift
from correlating response times from the perspective of end users to
analyzing the complete application system, and measuring service
requests running between the critical application components.
Traditional monitoring and application performance management tools
from the big vendors were not designed to work in concert across the
range of environments that support today's distributed applications:
physical, virtual or cloud.
Application management is the new war in virtualization, and it will be
here in full force in 2010. Application service management (ASM) -
which enables IT support teams to completely and visually monitor the
connections and components that comprise an application system - is a
key issue in this fight as it redefines what application performance
management should be. ASM will bring applications and infrastructure
ever closer to the business goals they support. And by extension, it
will improve overall business performance.
Datatrend's IT Optimization offerings can assist companies in their
efforts to optimize application management. One solution we offer is
BlueStripe's FactFinder, the only application management software
solution that delivers 24/7 application monitoring and true application
performance by measuring all real application service requests,
hop-by-hop, across the entire application system. Since applications
span multiple technology silos, FactFinder delivers the most accurate
application service level performance by measuring each inter-process
service request across all application components, including service
requests that cross technology silos. FactFinder provides a new
perspective on application performance in a matter of minutes -
requiring zero configuration, zero maintenance and zero coding on your
part.
For more information on BlueStripe FactFinder and other Datatrend IT Optimization offerings, click here.
[ back to top ]
|
|
Tech Tips
|
Want to learn more about Xen? Click the link below to take advantage of this month's Tech Tip pearl of wisdom:
Installing and Using Xen - Xen Virtualization Tutorial
This new, updated Xen tutorial is written by Zach Shepherd and Wenjin Hu, originally derived from materials written by Todd Deshane and Patrick F. Wilbur.
It was presented numerous times at Xen Summit in various locations across the United States.
Its purpose is to teach the installation, configuration, and use of the
free, open-source Xen virtual machine monitor (hypervisor), as well as
best practices. |
|
|
|
|
