![](http://welcome.hp-ww.com/img/s.gif) |
» |
|
|
![Content starts here](http://welcome.hp-ww.com/img/s.gif) |
![](http://welcome.hp-ww.com/img/s.gif) |
|
![](http://welcome.hp-ww.com/img/s.gif) |
![](http://welcome.hp-ww.com/img/s.gif) |
We will create a software platform that can virtualize a
variety of Alpha and VAX hardware. We want to be able to emulate enough
different systems to provide viable alternatives to any existing hardware
configuration, including multi-CPU systems.
Our emulators will support OpenVMS (VAX and Alpha
emulators) and Digital UNIX/Tru64 UNIX (Alpha emulators only) as operating
systems running on top of the virtual hardware.
Our emulators will be hosted on Integrity servers running
OpenVMS and Proliant servers running Windows. We will consider supporting a
Linux version of the product at a later stage.
When run on OpenVMS/Integrity as the host platform, our VAX
and Alpha emulators running OpenVMS will offer the same high-availability
features that real VAX and Alpha systems running OpenVMS have to offer.
In the future, we will explore the possibility of coupling
our emulators with hardware bus support to enable the use of the emulator with
custom hardware interfaces. It would be conceivable to see one of our VAX
emulators with an attached Q-Bus or XMI card cage used for replacement of
factory automation systems.
We are targeting a production release for a first Alpha
emulator in early 2010.
|
![](http://welcome.hp-ww.com/img/s.gif) |
![](http://welcome.hp-ww.com/img/s.gif) |
|
![](http://welcome.hp-ww.com/img/s.gif) |
![](http://welcome.hp-ww.com/img/s.gif) |
This section provides a high-level overview of the
emulator and shows how the various bits and pieces fit together.
Component hierarchy
The easiest way to think of an
emulator is to think of it as a piece of hardware because that is what it acts
like to the operating system and other software running on top of it.
Like the real hardware, the emulator consists of modules (components) that
interact with each other. Most of the components correspond directly to
physical hardware components. Components have a parent-child relationship to
each other. Child components are usually connected to their parent through a
virtual bus. For example, disk components are children of a disk controller
component, and PCI device components are children of a PCI controller.
Emulator component
This
abstraction poses a problem at the top-level of the emulated system. Most
systems have a top-level bus that has no real controller to act as its parent.
For example, in the AlphaServer ES40, the top-level bus consists of the D-chips
that connect the CPU’s to the C-chips (system chipset), the P-Chips (PCI
controllers), and main memory. Therefore, the decision was made to create both
a VAX emulator component and an Alpha emulator component. These emulator
components act as the controller for the top-level bus. For ease of
implementation, these components also include main memory.
![](star4.JPG)
Master Control Program (MCP)
Finally,
different emulators and components like networks need to be tied together. This
is accomplished through the master control program component. (Are there any
fans of either Burroughs B5000 mainframes or the movie Tron out there?)
The
following (simplified) image shows the components used to
emulate an AlphaServer 400 and a MicroVAX, interconnected through an Ethernet
network that is also connected to the outside world through one of the host
system’s network interfaces. The yellow components are those that interact with
the outside world. The red line indicates the “extra” parent-child relationship
between the network interface cards and the network top-level component.
Virtualization layer
From the beginning, the emulator
was written to be very flexible. We first created a framework to be used for
writing different emulators. All forthcoming VAX and Alpha emulators will use
this common framework. Into this framework, we incorporated all of the
functions that all or most emulators will be likely to need, such as:
-
Functions for configuring the
emulator: instantiation, configuration, and connecting together of all
emulator components;
-
Functions for controlling the
emulator: structured, sequenced discovery, initialization, starting and
stopping of all emulator components;
-
Emulated Ethernet connectivity
between emulators;
-
Common interfaces to the
outside world for networking, hard-disk emulation and I/O components: for
example, communications ports provide the ability to communicate through a
telnet session or a physical serial port. This way, this functionality can
be shared by any emulated communications port without requiring additional
effort, simplifying the emulation environment and providing a more
consistent user experience;
-
Hiding differences between
different host systems from the emulator components, so the same emulator
will run on both OpenVMS and Windows;
-
Support
for making use of multi-CPU or multi-core host systems by threading;
-
Emulator
licensing and protection.
In short, these are all the
emulator functions that are not directly related to the bits, bytes and
registers of the emulated hardware. We’ve named this framework the
“Virtualization Layer” because it creates a complete virtual environment for the
individual emulators.
Hardware Platform Abstraction Layer
As we want our emulators to run on
both Windows on Proliant servers and OpenVMS on Integrity servers, we were
confronted with the fact that Windows and VMS behave differently. To avoid
having to write platform-specific code for each emulator component, we
implemented an abstraction layer as part of the virtualization layer that hides
these differences from the rest of our code. These differences are mainly in the
following areas:
-
Threading and locking. On VMS,
we use the Pthreads library; on Windows, we use the Windows API.
-
Physical device access. On
VMS, we use QIO’s; on Windows, we use various API’s.
-
Timekeeping
Putting all platform-dependent
code in one place helps us to keep our code base clean.
|
![](http://welcome.hp-ww.com/img/s.gif) |
![](http://welcome.hp-ww.com/img/s.gif) |
|
![](http://welcome.hp-ww.com/img/s.gif) |
![](http://welcome.hp-ww.com/img/s.gif) |
The emulator makes extensive use
of OOP, particularly of the features offered by the C++ language. While C and
C++ are reviled by some for their perceived cryptic nature (although there is no
rule that says C or C++ code has to be cryptic), they are commonly considered to
be the languages of choice for low-level, portable programming found in
operating systems, device drivers, and emulators. C and C++ give programmers a
level of control over the bits and bytes of their code few other high-level
languages offer, and C and C++ compilers that produce blazingly fast code are
available for virtually any platform.
The emulator makes extensive use
of OOP, particularly of the features offered by the C++ language. While C and
C++ are reviled by some for their perceived cryptic nature (although there is no
rule that says C or C++ code has to be cryptic), they are commonly considered to
be the languages of choice for low-level, portable programming found in
operating systems, device drivers, and emulators. C and C++ give programmers a
level of control over the bits and bytes of their code few other high-level
languages offer, and C and C++ compilers that produce blazingly fast code are
available for virtually any platform.
Classes
All components
are implemented as classes. That means that a class has been designed for each
different kind of emulated component. For instance, if a RQDX3 controller needs
to be emulated, a RQDX3 class will be written. Once the class exists, the
emulator can create as many instances of that class as required. For example,
to emulate a VAX with three RQDX3 controllers, three instances of the RQDX3
class would be generated.
Inheritance
The RQDX3
controller needs to be able to interface with the Q-bus controller and vice
versa. The mechanisms involved are the same for all Q-bus devices; the way the
RQDX3 communicates with the Q-bus controller is no different than the way a
DELQA network interface communicates with it. Because of this shared behaviour,
all Q-Bus components share a common base class, the Q-Bus Device base class.
This way, the Q-Bus controller can address each of its child components as Q-Bus
Devices, rather than as individual types of interface. This takes full
advantage of the power of inheritance, a defining feature of OOP.
Component Base Class
The Q-Bus
Device class in turn has the Component class as its base class. The Component
class is part of the framework, and provides for such basic emulator-wide
functions as naming components, creating parent-child relationships,
initializing, stopping and starting the emulator.
Multiple Inheritance
It gets
trickier though. Besides being a Q-Bus device, the RQDX3 is also a disk
controller. As not all disk controllers are Q-Bus devices (for example, the
KZPAA SCSI disk controller is a PCI device), the disk controller base class
can’t have the Q-Bus device class as its parent. So, the RQDX3 needs to have
both the Q-Bus device class and the disk controller class as its base classes.
This is called multiple-inheritance.
In the case of
the KZPAA SCSI controller, it is even more complicated; it inherits from PCI
device, Disk controller, and SCSI device classes. The following diagram
illustrates this:
![](star2.JPG) |
![](http://welcome.hp-ww.com/img/s.gif) |
|