Improved Service Delivery in Network Infrastructure

Through Real-Time Virtualization™ on Intel® Virtualization Technology

Telecommunication Equipment Manufacturers (TEM’s) must deliver to Operators (Telco’s) new
generations of equipment, which satisfy more and more drastic requirements including reduced
Total Cost of Ownership, through both CapEx and OpEx reductions with increased processing
capabilities to serve more calls, and users per seconds in a reduced floor space, while
consuming less power. Supporting end to end added-value services is also an increasing
demand. These requirements are encountered in all areas of Network Infrastructure including the
Transport Network as well as new IP Multimedia Subsystems (IMS) and IP-TV services.

Historically this has been achieved by increasing the processing power of the processor and the
equipment density through cPCI, ATCA (Advanced Telecom Computing Architecture) and Micro-
ATCA form factors. Higher density and increased processing power at lower power consumption
can now be achieved through multi-core processors. This tends to be a disruptive evolution since
it discards the uni-processor paradigm used so far and deeply reflected as an implicit assumption
made by many legacy software stacks.

This whitepaper describes some of the technical and market challenges faced by network
equipment designers and explains how a real-time virtualization software solution combined with
Intel® Virtualization Technology simplifies many of these challenges. It discusses how other
solutions such as migrating legacy application software to Symmetric Multi-Processing (SMP)
platforms could be used to satisfy network equipment requirements but are determined not to be
sufficient when compared to using real-time virtualization technology on Intel® Virtualization
Technology.

© 2006 VirtualLogix , the VirtualLogix Logo, VirtualLogix VLX and Real-Time Virtualization are registered trademarks or trademarks of
VirtualLogix . Linux is a registered trademark of Linus Torvalds. All other names mentioned are trademarks, registered trademarks or service
marks of their respective owners.
Real-Time Virtualization™
Main Challenges Faced by Network Equipment Designers
For a given traffic capability, fewer
boards leads to reduced CapEx
and OpEx.
However multi-core processors
raise new challenges to existing
software stacks designed for uni-
processor configurations.
Reusing proven and state-of-the-
art IP related protocol in its original
runtime environment eliminates
lengthy adaptation, debug and
validation phases.
http://www.virtuallogix .com
Reduce Total Cost of Ownership (TCO) for
Operators.
More powerful hardware permits to process more calls on the same
board factor and to support more end-users, and more traffic, hence
offering the ability to reduce the number of required boards to handle
a given traffic capability. It also implies reduced power consumption
and possibly less cooling. Nowadays, increased power is no longer
achieved by increased processor speed but through multi-core
processors and SMP blades. This technology shift sets new
constraints on software such as the ability to run on SMP hardware
as well as scalability issues when the number of cores increases.
Each mobile phone operator needs to deploy many antennas to
provide an attractive geographical coverage. Antennas require Base
Transceiver Station (BTS) to operate. In some configurations, it may
be economically wise for an operator to share the cost of a common
BTS with other operators rather than each operator paying for their
own private BTS. Sharing a BTS between multiple operators
requires the ability to securely host different Telco’s on a unique
BTS and mandates that each operator retains full control and
management of their share. In other words, the software used by the
different operators must be securely isolated from each other, rules
to access shared or common resources must be strictly enforced,
and potential misbehavior of a share must not compromise the
service provided by other shares.
Timely Accommodate Evolutions to Provide Added
Value Services to the End-User with the expectation
of increased Average Revenue Per User (ARPU).
IP Network Introduction:
While Plain Old Telephone System (POTS) has been stable for
years, its convergence with IP networks raises some new issues. IP
family protocols are developed in general purpose operating
systems such as Linux whereas telephony signaling software has
mainly been based on dedicated real-time operating systems
(RTOS). Convergence and Time To Market pressures mandate the
ability to easily follow evolutions of IP related protocols without
having the burden of integrating IP stacks in legacy OS’s.
Conversely, one needs to leverage existing signaling protocols
implemented in legacy OS’s.
2(12) JL/TR-06-42.3
Real-Time Virtualization™

Radio Protocols Evolution:

The mobile phone infrastructure has been through major evolutions
in only a few years: from GSM to GPRS to UMTS to HSDPA and
beyond (LTE, 4G…), to provide higher bandwidth to terminals with
the goal to deliver quality multimedia streaming and other value-
added services to end-users. Newly designed equipment must
leverage the existing software, at no or very low cost. They should
also be able to accommodate future evolutions of the network
infrastructure.

Fixed-Mobile Convergence (FMC):

The convergence of fixed and mobile phones and the ability of
mobile phones to communicate over radio telephony protocols as
well as WiFi and WiMax networks, increase the complexity of the
network architecture and individual equipment both at the hardware
and software levels. Easily configurable systems help coping with
this increasing complexity while minimizing costs.

Providing Value-Added Services:

New end-user services delivered on the users’ terminal not only
require powerful content delivery servers, but also require support
from the communication network so that appropriate Quality of
Service (QoS) and security levels subscribed by the end-user are
correctly supported. Subscriber device management also require
some tight integration with the communication network. Richer and
more flexible software environments help solve this requirement.

Some of the above requirements are better fulfilled by re-using

legacy software with real-time constraints while others are better
accomplished on general purpose operating systems such as Linux
or Windows. Hence, there is a need to accommodate these different
software environments on the new hardware with higher processing
power and density mandated by the TCO reduction requirement.

Delivering in a Shortened Time To Market (TTM)

Window.
Given a new hardware platform providing increased processing
Reusing an unmodified legacy
capabilities and higher density, TTM is mostly dictated by software
protocol stack in its original runtime
development, adaptation and validation. Reusing existing and
environment eliminates lengthy
validated software within their original environment – whether it be a
debug and validation and shortens
real-time, legacy, Linux, or Windows environment is therefore
the Time To Market window.
critical.

http://www.virtuallogix .com 3(12) JL/TR-06-42.3

Real-Time Virtualization™
To mitigate risks introduced by
higher processing density,
additional High-Availability
mechanisms must be introduced.
Dilemma faced by Network Equipment designers:
“How to increase network equipment density while leveraging existing software stacks from
telecommunication networks as well as IP-based services, without compromising availability and
security ?”
http://www.virtuallogix .com
To effectively keep TTM as short as possible, no modification must
be required to the environment being re-used. This helps insuring
QoS constraints previously met are still supported, that the validation
phase can go seamlessly and market deployment will be reached
sooner.
Increase Dependability and Security.
Telecommunication networks have to be highly reliable. Operators
may have to pay financial penalties in case of lasting failures.
Hence, manufacturers must provide equipment –both hardware and
software- that sustain at least five 9’s (99,999 %) availability which
translates into less than 5 minutes of planned and unplanned outage
per year.
Fulfillment of the above mentioned capabilities must be done without
decreasing the availability of systems currently achieved. More
traffic is at risk in case of failure of a board due to increased
processing density. Software must be designed to limit the
consequences of a failure. Actually, wherever possible, availability
must be increased by permitting better fault isolation and detection,
faster failure recovery and a more flexible OAM&P (Operation,
Administration, Maintenance and Provisioning) permitting more
flexible upgrades.
Security is also a major concern in such environments. Introduction
of IP-based services and combination of various communication
technologies creates new threats. The increased density implies that
more calls and end-users are put at risk in case of a security breach.
Again, software must be designed to limit such risks.
4(12) JL/TR-06-42.3
Real-Time Virtualization™
An ideal design should enable multiple unmodified existing software stacks to
simultaneously run on a multi-core architecture, while sharing hardware
resources between securely isolated partitions.
General purpose OS’s have
required many years of effort to be
well-tuned for SMP configurations.
Scaling the performance in line
with the number of processors is
not an obvious task when the
number of processors increases.
Being able to run unmodified
existing operating systems
independently of one another is a
much more attractive solution.
http://www.virtuallogix .com
Multi-core processors deliver increased processing
power while minimizing physical footprint and power
consumption.
The aggregation of multiple cores within a single processor is an
effective way to provide higher processing power while reducing the
power consumption. This mandates that I/O peripherals that were
previously accessed from different boards can now be accessed
from a single one. The power required to support a board with a
quad-core processor is likely to be much smaller than the power
needed to handle four uni-processor boards. Overall heat dissipation
will also be lowered.
Legacy software must run on multi-core processors
without modifications.
Adapting legacy and real-time software initially designed for single–
core architectures to multi-core SMP architectures is a lengthy and
difficult process. Scheduling policies used in real-time environments
assume a single core processor. For TTM and cost reasons, an
ideal solution must allow real-time software to be re-used without
any modification to enable a controlled path and time to validation
and market availability. This requires hiding the multi-core properties
of the hardware and providing the software with the usual single-
core hardware properties it has been designed for.
A similar need may be encountered for general purpose operating
systems even when already adapted to SMP architectures.
Consolidating systems previously packaged as independent sets of
hardware and software means aggregating several software stacks
within a single environment. This is a painful and lengthy task that
requires merging different system versions into a single one as well
as integration and re-validation. This increases TTM and the
development costs of the product.
We will now examine some implementation alternatives:
Adapt software to multi-core environment,
• Hide multi-core through paravirtualization,
• Rely on fully transparent Real-Time Virtualization.
5(12) JL/TR-06-42.3
Real-Time Virtualization™

Implementation Alternatives

Adapt Software to Multi-core

Since lowering the Total Cost of Ownership is best achieved by
homogeneous multi-core SMP processors, software could be
adapted to such processing capabilities. This approach is attractive
because, once adapted, software runs on the multi-core processor
taking full benefit of the parallelization provided by the hardware.

However, this approach has some major drawbacks:

• Porting operating systems from uni-processor to SMP hardware

architecture is a complex task requiring a lot of tedious analysis
and tuning of the system to reach the expected level of
performance, i.e. getting close to twice the performance of a
single core out of a dual core, and scaling linearly as the number
of cores increases.
• Worse, this issue is not only related to the operating system;
application software is impacted as well by a move from uni-
processor, real-time software in particular. The scheduling
properties of a real-time environment are often based on FIFO
policies. Moving to multi-cores and thereby enabling lower priority
tasks to run in parallel with higher priority tasks will very likely to
break many synchronization assumptions. Hence applications
need to be adapted and fully validated.
• Consolidation of multiple independent systems running on
different blades into a single SMP system may remove some
configuration flexibility. Some low-end hardware configurations
may still require uni-processor blades with different kinds of
operating systems (some running a RTOS, some running a
general purpose OS), or may run different OS distributions
(libraries, tools, sets of patches) of the same system. Moving to a
single SMP operating system environment would eliminate such
configuration flexibility which may present a tough choice at
configuration time.
• Running four instances of a system environment on four uni-
processor blades physically partitions the system. This has some
advantages: failure of one of the system does not impact the
other ones. Breaking the security of one of the system does not
automatically give control over the three other blades. Such
advantages are lost when moving to a single software system
environment on a multi-core platform.

http://www.virtuallogix .com 6(12) JL/TR-06-42.3

Real-Time Virtualization™

Paravirtualization
Paravirtualization presents the guest system with a view of the
underlying hardware that is controlled by a Virtual Machine Monitor.
Each guest system runs isolated in its own partition without being
able to interfere with systems running in other partitions on the same
platform. In addition:

• Real-time systems can be presented with a uni-processor view of

the world which avoids the burden of adapting their application
stacks. One such real-time environment can run on each core of
the processor.
• One may run different systems or system versions side by side
on the same platform without having to go through tedious and
complex system integration tasks.

• By partitioning the physical resources (including the multi-core

processor) into different software environments, there is no single
domain of failure at the software level. Similarly, from a security
stand-point there are multiple security domains, breaking one
such domain does not provide full control over the other domains.
However, some aspects of this paravirtualization approach could be
improved:

• While it preserves good performance levels, it is at the cost of

adapting the Hardware Abstraction Layer of the guest system to
the Virtual Machine Monitor interface. Although this is a well-
understood task with limited scope and effort, it introduces delays
in the adaptation of the OS and may require partially revalidating
some of the software.
• Paravirtualization is usually achieved by changing the
environment in which the guest system runs natively: changing
the ring of execution on Intel x86 architectures (ring
compression), changing the available address space (address
space compression). This requires some attention.

Paravirtualization is the best solution when targeted processors do

not provide support for virtualization. This is the case when some
instructions accessing machine state can be silently executed in
user mode rather than causing an exception. This was the case for
x86 architecture prior to the introduction of Intel Virtualization
Technology processors.

With such support of transparent virtualization, an even better

solution can be applied.

http://www.virtuallogix .com 7(12) JL/TR-06-42.3

Real-Time Virtualization™

VirtualLogix VLX Real-Time Virtualization Technology on Intel® Virtualization

Technology

Full Transparent Real-Time Virtualization

The combination of fully virtualizable processors and an appropriate
Virtual Machine Monitor fulfils all of the requirements we have
established so far.

New processors such as those based on the Intel Virtualization

Technology enable Virtual Machine Monitors to transparently run
multiple unmodified guest OS’s either on a single processor or on
multi-core processors.

The unique combination of Intel Virtualization Technology and

VirtualLogix VLX satisfies all the requirements we have listed so far
without any of the drawbacks of other solutions that have been
mentioned.

The most important benefits are as follows:

• Consolidation: Multiple RTOS instances may run in parallel on

an SMP multi-core processor without disrupting the behavior of
the supported applications.
• Binary Legacy Reuse: No guest OS modifications are required
to achieve such a configuration, therefore enabling the shortest
Time To Market, and minimum development costs.
• Mix RTOS and General Purpose OS (GPOS): One or more
GPOS such as Linux or Windows may be run on the multi-core
processors, either on a dedicated core or on multiple cores
shared with other OS’s (even with one or several RTOS’s).
• Device Assignment: Devices can be dedicated to a given
partition or efficiently and securely shared between different
guest operating systems.
• Increased Availability: Guest systems may be transparently
monitored and depending on the configuration may be
automatically restarted in case of failure. Hence overall
dependability of the system is increased without having to modify
guest operating systems.

http://www.virtuallogix .com 8(12) JL/TR-06-42.3

Real-Time Virtualization™
VirtualLogix VLX manages key
resources, through partitioning
and virtualization, isolating guest
OS's from the hardware.
http://www.virtuallogix .com
A Virtual Machine Monitor, VirtualLogix VLX, runs in the “root” mode
of processors with the Intel Virtualization Technology and manages
key system resources to isolate guest OS's from the underlying
hardware.
Typically, physical memory is partitioned between guest OS's based
on configuration description, while the CPU, FPU, MMU or other
system components, such as the real-time clock and interrupt
controller, are virtualized by VirtualLogix VLX.
One or more guest OS’s may share the same core within a multi-
core processor. If multiple guest OS’s run on the same core,
VirtualLogix VLX schedules the guest OS’s in such a way that real-
time properties of RTOS’s are maintained.
This approach enables the VirtualLogix VLX virtualization technology
to be applied to embedded and real-time systems.
Partitioning
Physical memory is partitioned, and each memory partition is
allocated to a given guest OS. Thus, each guest OS may use its
own native memory management mechanisms and policies, without
interfering with other guest OS's. Memory partitions are protected
and isolated from each others so that no OS can read or write into
the memory partition of another OS. This approach allows OS’s that
make use of the MMU, e.g. Linux or Windows CE, and those that do
not, e.g. Nucleus or VxWorks, to run cooperatively on a single core.
I/O devices used by a single guest OS are assigned exclusively to
that guest OS. Native device drivers can thus be re-used with no
modification.
Virtualization
Hardware resources required by several guest OS’s, such as the
CPU and real-time clock, are virtualized so that they can be shared
between all guest OS's that need to access them.
The Intel Virtualization Technology architecture enables to
transparently virtualize the processor avoiding the need to modify
guest OS’s as required when paravirtualization is the only solution.
No dynamic binary translation is even required to have guest OS’s
run in a partition on top of VirtualLogix VLX for processors
supporting Intel Virtualization Technology.
9(12) JL/TR-06-42.3
Real-Time Virtualization™

VirtualLogix VLX virtualizes the CPU, FPU and MMU. The CPU is
shared among guest OS’s based on scheduling policies that
guarantee that real-time guest OS’s (and their applications) are
assigned the highest priorities and can thus maintain their real-time
guarantees. Once a guest OS has been granted CPU access, the
OS uses its native policies to schedule its own applications and
services.

The MMU is virtualized so that each guest OS may use it for its own
purposes. As discussed above, usage of the MMU by one guest OS
is independent of the usage of the MMU by another guest OS.

Device Virtualization
Running different OS's simultaneously on the same processor is
only one part of the problem. The guest OS's also need to
communicate and share devices or other hardware resources.

A typical OS supports multiple processes, provides them with

memory allocation and scheduling policies, and offers services such
as synchronization, shared access to system and network interfaces
and inter-process communication.

Similarly, the VirtualLogix VLX virtualization technology provides

each guest OS with synchronization (a cross-interrupt mechanism),
shared access to devices such as disk controllers, network
interfaces, serial lines and inter-OS communication mechanisms,
through virtual devices (virtual Ethernet or virtual UART).

I/O devices fall in one of the three following categories:

• Dedicated I/O devices are used by only one guest OS, and
require no driver modification
• Shared I/O devices may be used by more than one OS; they
require specific cooperating drivers to permit physical I/O device
sharing between various guest OS’. VirtualLogix has developed a
portfolio of such drivers, such as Ethernet, audio, frame buffer,
disks, UART, modem, keyboard, etc.
• Virtual devices do not correspond to a physical device, but are
used for inter-OS communications. Again, VirtualLogix has
several such virtual devices available, such as UART, Ethernet,
etc.

http://www.virtuallogix .com 10(12) JL/TR-06-42.3

Real-Time Virtualization™

VirtualLogix VLX at Work on Processors Supporting Intel® Virtualization

Technology

VirtualLogix’s customers are using VirtualLogix VLX virtualization

technology to consolidate blades and processors in their softswitch
equipments. The unique combination of Intel Virtualization
Technology and VirtualLogix VLX enables the consolidation of such
equipment by using blades equipped with SMP multi-core
processors.

Not only does this approach enable several instances of a legacy

RTOS to run on a multi-core processor without having to modify
them, but it also permits to run Linux system instances on the
equipment without having to specifically provision dedicated blades
or processors. Hence, this brings a full new set of capabilities to
platforms historically running legacy systems. This opens up a
controlled migration path to Linux.

Let’s illustrate the capabilities offered by the combination of

VirtualLogix VLX and an Intel Virtualization Technology processor,
on a Home Location Register (HLR) demonstrator- a database
server containing details of mobile phone subscribers.

On an ATCA Intel NetStructure® MPCBL0040 board equipped with

two Dual-Core Low Voltage Intel® Xeon processors, VirtualLogix
VLX enables multiple RTOS and a Linux instance to run
simultaneously on the same board.

RTOS and their application stacks run without modification; each of

them behaves as if it were running on a uni-processor machine.
Setting up such a system is just a matter of configuration to dedicate
the appropriate physical resources to each partition.

Since each OS is independent of the others and has a dedicated

core (except the one sharing its processor with Linux), there is no
noticeable synchronization overhead and the system delivers close
to four times the processing power provided by a single core
processor.

As a result, four databases can be accessed simultaneously. This

allows optimal usage of the high-performance multi-core processors
without compromising original HLR application features.

The four unmodified databases can communicate with a network file

system through a standard Linux (Red Hat Enterprise Linux ES 4)
running on one of the cores.

http://www.virtuallogix .com 11(12) JL/TR-06-42.3

Real-Time Virtualization™
VirtualLogix
292 Gibralter Drive, Bldg B2
Sunnyvale, CA 94089, USA
Tel +1 408 954 7355
Fax +1 408 432 7235
http://www.virtuallogix .com
Independence of the systems and their database servers implies
that the failure of one of the four systems does not prevent the three
remaining systems to carry on their duty.
Summary
The VirtualLogix VLX virtualization technology combined with Intel
Virtualization Technology solves many of the challenges of network
infrastructure equipment and enables manufacturers to:
• Design equipment based on multi-core processors, allowing
providing higher processing density while lowering the Bill of
Material and the Total Cost of Ownership.
• Consolidate multiple unmodified real-time software stacks on a
single multi-core processor by simply configuring the system
while maintaining their QoS properties.
• Run legacy, real-time and general purpose software without
complex and lengthy redesign and integration.
• Consolidate different versions of general purpose software stack
on the same multi-core processor without having to go through a
software environment integration process.
• Optimize the system to deliver the requisite product features and
performance.
12(12) JL/TR-06-42.3