System Specifications

---

Recommendations

In order to make our recommendations to the company, I should point out that each system has its advantages and disadvantages to our specified requirements. These are in order:

Sun Microsystems W2100z

The Sun System Utilizes the exceptional performance of the AMD Opteron Processor, and easy migration to 64-Bit operating systems. The PC3200 Memory running at 12.8GB per second offers improved speed and computational power, exactly what we require to run the high specification graphics that we produce. The NVIDIA FX3000 Accelerator will enhance our rendering procedures, cutting current rendering times by a third. The only disadvantage I can see is the fact that of the three models we have looked at this is the highest priced.

HP J6750

The HP J6750 is a RSIC (Reduced Instruction set Computer), and although primarily it is designed for use in a UNIX environment, and it’s relatively low processor speed of just 875Mhz, I believe that it would be possible for us to adapt it to a Microsoft or Linux based Operating system, in order to permit us to carry on using our current rendering software. IBM’s Microelectronics Division manufactures the PA-8700s under an OEM agreement. It has 1.5Mb of data cache and 0.75Mb of instruction cache on chip, a 50 per cent increase in cache sizes compared to the PA-8600. The graphics accelerator would also need replacing with an NVIDIA card as the standard card is not acceptable to our requirements.

HP XW8200

The XW8200 is an excellent all-round machine specification, it comes with a wide range of Intel Xeon Processor (see Appendix C) from 2.8 to 3.6 GHz with 1 MB L2 cache, and for our requirements of high-resolution graphics, the Video Accelerator card is exceptional value for money with high speed rendering and high resolution display area, will perform all the functions that we demand. The system also features Intel Extended Memory 64, which greatly increases the addressable memory area. The PCI-Express (x16) gives 4 times more bandwidth than the AGP x8, which again would give exceptional processing power for our rendering. (See Appendix E) for more details about PCI-Express.

Conclusion of Recommendations

It is of our opinion that we should invest in the HP XW8200 as it offers us the best in affordability and productivity, what we save on the hardware we can invest in more elaborate software to take full advantage of the functionality and power that the HP XW8200 provides.

        

---

Comparisons of Systems

Input/Output Technical Guide

Introduction to Technical Guide

The following reference information hereafter is a technical guide, requested by the Director of IT services to explain how a selection of Input/Output ports in unspecified computer architecture works. Inside there is information on the following ports:

• Centronics (Parallel)
• RS232 (Serial)
• USB 2.0
• SCSI  (see Appendix D)
• Firewire (see Appendix D)

For each, there will be a brief overview of the system, followed by a more technical description, containing data-flow and mechanical and electrical information.

A glossary of terms can be found at the back of this manual to enable you to understand the terminology used.

---

Centronics (Parallel) Interface

---

Serial Interface

The Serial port is considered to be one of the most basic external connections to a computer; the serial port has been an integral part of most computers for more than 20 years. Although many of the newer systems have done away with the serial port completely in favour of USB connections, most modems still use the serial port, as do some printers, PDAs and digital cameras. Few computers have more than two serial ports.

The term serial comes from the fact that a serial port "serializes" data. That is, it takes a byte of data and transmits the 8 bits in the byte one at a time. The advantage is that a serial port needs only one wire to transmit the 8 bits (while a parallel port needs 8). The disadvantage is that it takes 8 times longer to transmit the data than it would if there were 8 wires. Serial ports lower cable costs and make cables smaller.

Before each byte of data, a serial port sends a start bit, which is a single bit with a value of 0. After each byte of data, it sends a stop bit to signal that the byte is complete. It may also send a parity bit.

Serial ports, also called a COM port (which stands for communication), are bi-directional. Bi-directional communication allows each device to receive data as well as transmit it. Serial devices use different pins to receive and transmit data -- using the same pins would limit communication to half-duplex, meaning that information could only travel in one direction at a time. Using different pins allows for full-duplex communication, in which information can travel in both directions at once.

Serial ports rely on a special controller chip, the Universal Asynchronous Receiver/Transmitter (UART), to function properly. The UART chip takes the parallel output of the computer's system bus and transforms it into serial form for transmission through the serial port. In order to function faster, most UART chips have a built-in buffer of anywhere from 16 to 64 kilobytes. This buffer allows the chip to cache data coming in from the system bus while it is processing data going out to the serial port. While most standard serial ports have a maximum transfer rate of 115 Kbps (kilobits per second), high speed serial ports, such as Enhanced Serial Port (ESP) and Super Enhanced Serial Port (Super ESP), can reach data transfer rates of 460 Kbps.

An important aspect of serial communications is the concept of flow control. This is the ability of one device to tell another device to stop sending data for a while. The commands Request to Send (RTS), Clear To Send (CTS), Data Terminal Ready (DTR) and Data Set Ready (DSR) are used to enable flow control.

For example if you have a modem that communicates at 56 Kbps. The serial connection between your computer and your modem transmits at 115 Kbps, which is over twice as fast. This means that the modem is getting more data coming from the computer than it can transmit over the phone line. Even if the modem has a 128K buffer to store data in, it will still quickly run out of buffer space and be unable to function properly with all that data streaming in.

With flow control, the modem can stop the flow of data from the computer before it overruns the modem's buffer. The computer is constantly sending a signal on the Request to send pin, and checking for a signal on the Clear to send pin. If there is no clear to send response, the computer stops sending data, waiting for the Clear to send before it resumes. This allows the modem to keep the flow of data running smoothly.

Again although serial is used by such devices as modems, mice, PDA’s and Data capture devices, it is still slower than USB and is also prone to bottlenecks, as data is transferred. For this reason we again recommend that all of these serial devices be upgraded to USB.

USB Interface

Nearly every computer that is bough today has a USB (Universal Serial Bus) interface on it. Sometimes there is a few at the back, but more increasingly they are being added to the front of the system unit to allow for easy plugging and playing of peripheral such as digital cameras, printers, hard drives, PDA’s, Modems, Scanners and flash keys.

The idea behind USB was to relieve some of the headaches associated with installing multiple high-speed devices. In the past, it was only possible to associate a couple of devices with a parallel or serial port, for example a system may have had a printer and a scanner attached to the same parallel port, this caused the flow of data from the scanner to the system unit to become slow, but with USB there can be as many as 127 devices attached at any one time, with no loss in speed of data transfer. Connecting a USB device to a computer is simple -- you find the USB connector on the back of your machine and plug the USB connector into it.

The Universal Serial Bus has the following features:

• Up to 127 devices can connect to the host, either directly or by way of USB hubs.
• Individual USB cables can run as long as 5 meters; with hubs, devices can be up to 30 meters (six cables' worth) away from the host.
• A USB cable has two wires for power (+5 volts and ground) and a twisted pair of wires to carry the data.
• On the power wires, the computer can supply up to 500 milliamps of power at 5 volts.
• Low-power devices (such as mice) can draw their power directly from the bus. High-power devices (such as printers) have their own power supplies and draw minimal power from the bus. Hubs can have their own power supplies to provide power to devices connected to the hub.

Conclusions for Upgrade to USB2.0

After weighing up the advantages and disadvantages of the other interfaces we feel that the company must move with the times and upgrade all our serial and parallel peripherals to USB 2.0 as it is quite clear from the differences in Data transfer speeds that we would increase our productivity and reduce bottlenecks both on the individuals system and across the network.

In Appendix (D) we have displayed our findings into 2 other interfaces which we feel will compliment the existing IT strategy for the future, as we have recommended upgrading our external peripherals, it would seem pertinent to investigate how we can improve the data transfer speed of our storage media. Both Firewire and SCSI could give greater improvements to our current IDE Hard-Drives.

---

Appendix (A)

Direct Connect Architecture 

• Memory is directly connected to the CPU optimizing memory performance
• Input/Output is directly connected to the CPU for more balanced throughput and Input/Output
• CPUs are connected directly to CPUs allowing for more linear symmetrical multiprocessing

Integrated DDR DRAM Memory Controller 

• Changes the way the processor accesses main memory, resulting in increased bandwidth, reduced memory latencies, and increased processor performance
• Available memory bandwidth scales with the number of processors
• 128-bit wide integrated DDR DRAM memory controller capable of supporting up to eight (8) registered DDR Dimms per processor
• Available memory bandwidth up to 6.4 GB/s (with PC3200) per processor

Hypertransport Technology 

• Support of up to three (3) coherent Hypertransport links, providing up to 19.2 GB/s peak bandwidth per processor
• Up to 6.4 GB/s bandwidth per link providing sufficient bandwidth for supporting new interconnects including PCI-X, DDR, InfiniBand, and 10G Ethernet

Low-Power Processors 

• The AMD Opteron processor HE offers industry-leading performance per watt making it an ideal solution for cooler, quieter workstation designs.
• The AMD Opteron processor EE provides maximum Input/Output bandwidth currently available in a single-CPU controller.

---

Appendix (B)

PA-8700 Architecture (PA – Precision Architecture)

The following diagram shows the architecture behind the PA-8700 series processor.

Fig. 8 PA-8700 Processor Architecture

()

The Instruction words of the PA-8700 have a constant width of 32bits, which simplifies the instruction decoding logic. Instructions are issued directly in silicon, therefore microcode is not required. There are limited addressing modes so the most frequent of operations can be performed at a faster rate.

---

Appendix (C)

Fig. 9 Intel Xeon high-level Diagram

The execution resources of the Intel Xeon processor are shared between to logical processors; the Rapid Execution Engine and the Integrated Cache subsystem, which process the information simultaneously. The Fetch and Deliver Engine, gathers instructions from each of the logical processors in turn and sends these instructions to the Rapid Execution Engine.

At the Rapid Execution Engine, both sets of instructions are executed at the same time, taking instructions from the queue, although the instructions can be taken out of order. The Integrated Cache Subsystem delivers data at high speeds to the core processors, and because the subsystem is clocked at the same rate as the processor core, as faster processors are released the Cache Subsystem is also increased.

The Reorder and Retire block takes the instructions, that were operating out of order and puts them back into program order. The system bus provides up to 4.27GB/sec bandwidth, running at 533MHz.

 

---

Appendix (D)

IEEE1394 Firewire

FireWire was originally created by Apple and later standardized as IEEE-1394, actually preceded USB and had similar goals. Firewire is a serial connection that operates at very high data transfer speeds. The average speed of data transference is between 100 and 800Mbits per second, although it has been promoted that it will transfer data up to 3200Mbits per second. Firewire was originally intended for devices working with lots more data -- things like camcorders, DVD players and digital audio equipment. Firewire and USB share a number of characteristics and differ in some important ways. Here's a summary:

• Like USB, Firewire is a serial bus that uses twisted-pair wiring to move data around.
• However, while USB is limited to 12 megabits per second, Firewire currently handles up to 400 megabits per second.
• USB can handle 127 devices per bus, while Firewire handles only 16 on a single Firewire port.
• Both USB and Firewire support the concept of an isochronous device -- a device that needs a certain amount of bandwidth for streaming data. This mode is perfect for streaming audio and video data.
• Both USB and Firewire allow you to plug and unplug devices at any time.

SCSI – (Small Computer Systems Interface)

SCSI – pronounced “Scuzzy” is primarily a storage device interface, (although it can also be used for other peripherals such as scanners) for high-end business machines, although some home users like to use it, to increase their data access speeds.  

With a SCSI adapter you can connect up to 16 devices to one adapter, but the last device in the line has to be “terminated”; this means that at each end of the SCSI bus is closed, using a resistor circuit. If the bus were left open, electrical signals sent down the bus could reflect back and interfere with communication between SCSI devices and the SCSI controller. Only two terminators are used, one for each end of the SCSI bus

There are three main types of SCSI Interface, but quite a few variations of these exist (see Fig9)

. The standard types are:

• SCSI 1 – the original developed in 1986
• SCSI 2 – became the standard in 1994 and included the Common Command Set (CCS), which included 18 commands for the support of SCSI. Also included was the option to increase the clock speed from 5Mhz to 10 MHz, increase the bus speed from 8bits to 16 bits per second, and increase the number of connectable devices to 15
• SCSI 3 – released in 1995, is not accepted as the standard as it is continuously evolving

Fig. 11 SCSI Types Comparison Table

---

Appendix (E) Parallel Port Interface

Fig. 12 Parallel Port External View

Fig. 13 Parallel Port Pin Configuration

---

Appendix (F) PCI –Express System Architecture

The basic PCI Express design consists of a root complex (possibly a Northbridge sort of chipset device that connects the CPU/memory subsystem to I/O devices), switches (which internally appear to software as containing two or more logical PCI-to-PCI bridges to maintain compatibility), and various endpoint devices. Certain bridge devices, like PCI-to-PCI Express could also exist.

A key feature of PCI Express is the embedded clocking technique; it uses 8b/10b encoding which is encoded directly into the data stream as opposed to having the clock as a separate signal. The 8b/10b encoding requires 10 bits per character.

The PCI Express link is comprised of many lanes; each lane is made up of two different pairs of wires, one to transmit and one to receive data. The lanes scale from 2.5Gbits/sec in each direction, stretching to 10Gbits/sec in the future. Multiple lanes can be connected between devices. PCI Express is configured in 1x, 2x, 4x, 8x, 12x, 16, and 32x lanes widths.

PCI Express supports two types of interrupts, the older PCI INTx (where x= A, B, C, or D) legacy interrupt using an emulation technique, and the newer Message Signalled Interrupt (MSI) capability. MSI is optional in PCI 2.2/2.3 devices, but required as the native mode of PCI Express devices. The INTx emulation can signal interrupts to the host chipset. It's compatible with existing PCI-compatible driver and operating system software. It virtualizes PCI physical hard-wired interrupt signals by using an in-band signalling mechanism. PCI Express devices must support both the legacy INTx and MSI modes, and legacy devices will encapsulate the INTx interrupt information inside a PCI Express

MSI interrupts are Edge-Triggered and sent via memory write transactions. The MSI scheme is the desired method of interrupt propagation when using a packet protocol over a serial link, and MSI's are more effective in multi-processor systems as any device can issue interrupts to different hosts directly.

Fig. 14 PCI Express Connections to motherboard

---

References

Below is a list of references from where I have drawn the information and images in order to compile this document.

22nd October 2004

22nd October 2004

23rd October 2004

1st November 2004

1st November 2004

1st November 2004

 

1st November 2004

9th November 2004

20th October- 7th November (Image Searches for PCI Express, Parallel Ports, SCSI Terminator)

Information systems development: methodologies, techniques, and tools.

David Avison and Guy Fitzgerald

A+ Certification – Microsoft Press

Principals of Computer Hardware – Alan Clements

Fundamentals of Computer Architecture – Mark Burrell

PCI Express System Architecture – Don Anderson, Tom Shanley, Joe Winkles

Parallel Port Complete – Jan Axelson

USB Peripheral Design – John Koon

Universal Serial Bus Explained – Steven McDowell, Martin Seyer

---

Glossary