Beyond the limit - detail IA-64 architecture
Gong Yu Zhang Haitao
What is IA-64?
IA-64 is an Intel's 64-bit architecture that calculates the EPIC (ExPlicitly Parallel Instruction "technology based on dominant parallel instructions, eventually implementing the following objectives:
Overcoming limitations of traditional architecture;
Provide advanced floating point calculation performance;
Support 64-bit memory addressing;
Compatible with 32-bit systems, maintaining its original investment;
Supports growing high-end application workloads in e-commerce, enterprises, and critical calculations.
Editor's note: Since 1994, the IA-64-bit architecture has already passed the rainy history of nearly six years. Although it is ready to move, it is still half-faced, but no matter how it said, it has already got a lot. Manufacturers' support, IBM, Microsoft, Sun are actively introducing an operating system based on the IA-64 platform. Several transformations, the first generation of IA-64 is expected to be introduced in 2000. Before it introduced, let's have a comprehensive understanding of its overall characteristics!
Since 1993, Intel and its partner companies have launched a 486 system-based IA server, IA servers have experienced a 486 system, a Pentium Pro system, a PII system, a PIII system, a Xeon system and other stages, and the processing power of the processor system is substantially Improve, and the bus structure of the server system is always the IA-32-bit bus system.
Break through bottleneck
For traditional computer applications, database applications, file print services, IA-32 servers based on SMP multiprocessor bus systems can be easily cop. However, network technology, network application development has broken all this, and information transmission is no longer limited to a few points, but hundreds of points, but hundreds of thousands of points, even more, is "World Wide Web", on the network There are hundreds of megabytes of information and even more data traffic per second, which has higher requirements for the I / O performance, bus performance, and processing capabilities of the web server. Looking at the development of the IA server, it can be said that IA servers are technically, and the product has now come up to develop, breakthrough edges. After the IA-32 server has developed to the 8th Xeon server, the architecture has begun to become a bottleneck that is constrained by the server performance. First, the PCI channel bandwidth bottleneck is now the memory bus bandwidth bottleneck, the processor system expands the bottleneck. In summary, with the emergence of each generation of microprocessors, the performance of traditional structures gradually began to become outward, and in practical applications, the demand for those in parallel computing is gradually increased. Therefore, HP and Intel have been working together to develop IA-64 architectures since 1994, hoping to combine HP to combine HP's decade of work in the RISC field, improve performance on microprocessor levels To increase the parallelity of the instruction level.
Let's first look at the traditional structure to limit the performance of the processor.
1. The message is slow, and only a few instructions can be handled at a time.
2. Branch, that is, change the execution of the data stream in the program. The branch judge which instruction set is loaded, and when it is wrong, the entire process will delay.
3. Internal memory delay, that is, the time from the memory is retrieved. The internal memory delay delays the processor's operation to be idle until the data processing is accepted from memory to start operation. Because of the memory delay, you cannot keep up with the speed of the processor, so you need to start from the memory search data to ensure that the data can be transferred in time.
So how do IA-64 break through these restrictions?
In a parallel mechanism, the compiler can effectively organize the code and make the execution order clearer to enable the processor to perform instructions in a more efficient manner.
In order to solve the problem, the new IA-64 instruction set structure (ISA) uses a new concept - branch pre-judgment. It can remove excess branches to reduce errors. Strengthen the branch prediction function in the execution, improve the hit rate of branch prediction, thereby making the pipeline smoothly. For internal memory delays, dynamic execution techniques are used in the IA-64 architecture, that is, during the running process of the program, when a data may be discovered, the data is taken out in advance and stored in the register standby.
IA-64 unique features
EPIC is the most distinctive technology of IA-64, so-called EPIC technology, simply, the compiler first analyzes the dependencies between the instructions, and will combine the instructions (most 3 instructions) without dependencies (up to 3 instructions) into a group, then The completed command group is read into a group-in-set command group and executed by the built-in execution unit. EPIC technology can reduce the cost of the processor because the software has determined the allocation of each instruction, so it can improve the parallelism of the instruction using only simple hardware.
The main characteristics of IA-64 are manifested in several aspects.
First, the system memory addressing space of IA-64 is larger, can support 32GB of memory, and the maximum memory capacity of the IA-32 server is 16GB.
The processor addressor of IA-64 is more processing, the speed is faster, and the ITANIUM processor will start at least 1 GHz, and the secondary Cache is 2MB.
The 128-bit floating point calculation register of IA-64 system has greatly improved the floating point computing power of the system.
The IA-64 system will use InfiniBand-based bus structure, which is the core of the exchange system bus, which is the core, and two technologies of NGIO and FUTURE IO are one, so that the system bus, memory bus bandwidth and The I / O bus bandwidth will be greatly improved, the system bandwidth is above 2GB / s, and the current SMP IA-32 server system bandwidth is 1.06Gb / s, and the PCI bandwidth is generally 0.4Gb / s.
Second, IA-64 includes a range of built-in features to extend the normal operation time of the computer and reduce downtime. Especially the machine detection system (MCA) provides error recovery and error correction (ECC) capabilities in memory and data paths. It allows the IA-64 platform to recover from the error that leads to the failure of the system. ECC and corrective circuitry are automatically fixed to keep the system running status. In addition, the memory fault tolerance characteristics of the IA-64 chipset allows IT staff to reset memory without rebooting the program.
Finally, from the programming point of view, Intel has developed an assembler that is to help assemble programmers with efficient IA-64 assembly code and compilation synthesis program written by Intel and HP. In the 64-bit structure, programmers can easily determine which command in each cycle can be executed. A more efficient instruction-level parallel mechanism can be created based on a parallel mechanism and a large amount of floating point registers with many execution units. The register-based control instruction can reduce the number of packets, data, and control determination can reduce memory delay, thereby obtaining a higher speed. The new compilation technology uses these features, which is possible to improve the performance of the assembler to optimize performance.
Operating system ready to go
With a good hardware platform, we must consider the software. Provide system vendors that support 24 × 7 e-commerce will support IA-64-based servers. Many companies will also provide IA-64 systems with service level protocols that ensure availability. At the same time, e-commerce will also benefit from the operating system based on the IA-64 platform. Currently, there is no other platform for operating system bandwidth, which is only found in IA-64. It is now officially announced that there are MonteRey, Linux64, HPUX, Solaris, Win2000, etc., which support IA-64 platforms. In the platform, MonteRey is a particularly special system, which is the AIX system of SCO's Unixware plus IBM. IBM promises that AIX applications under all RISC platforms can be seamlessly transplanted to the IA-64 Monterey platform. From another perspective, some people have filed a unified UNIX camp, allowing the applications in each UNIX camp to transplant, so SCO and IBM launched MonteRey can be seen as a positive measures for unified UNIX camps.
Migration to 64-bit system
After the introduction of IA-64, people are worried that the most problem is probably the system's migration. For example, the computer products of HP are actually transferred from the PA-RISC system to the IA-64 system. In fact, in terms of migration, the manufacturers have already considered it. The main feature of the IA-64 architecture is compatibility with the IA-32 instruction set. At the IA-64 system environment, the processor can perform IA-32 and IA-64 instructions and define three special instructions and interrupts to centralize the IA-32 and IA-64 instructions. This is the IA-64 system provided with binary compatibility with the IA-32 instruction set, and IA-64 processors can run IA-32 applications in IA-32 applications that support IA-32 applications. Mix the IA-32 and IA-64 architectures to perform operational capacity.
HP's approach is that for IA-64 technology, in which a key part is to implement binary compatibility with existing HP systems. To make HP's PA-RISC / HP-UX and IA-32 / NT systems and workstations can be compatible with HP IA-64 systems, that is, HP users can implement smoothing conversion of the new architecture, and Do not require the entire system to complete the migration. HP and Intel's cooperation can make HP all HP-UX and NT applications can run without worrying under the IA-64 system.
For hardware investment protection, HP new IA-64 systems support both PA-RISC and IA-64 processors. As long as you install a IA-64 processor upgrade board, you can upgrade to the IA-64 microprocessor system.
In summary, the purpose of IA-64 is to integrate UNIX, NT and Linux into a single industry-standard hardware platform, providing customers with greater flexibility, so that they can come to the most flexible and most cost-saving way. Design their development environment.
Unaffected
For the server
Due to the changes in the system structure, the performance characteristics of the server have also changed significantly. From the underlying structure, since its underlying processor instructs pipe growth, the number of register bits is growing, and the calculated data bit is longer. In particular, the control bit is increased, and the judgment is also increased, so that the security of the data is more secure. This gives the server that information storage and processing can be completed in memory, which can carry more information, to improve response time, significantly increased work efficiency, double the IA-32 system, greatly strengthened. From an availability, the IA-64 architecture enables the server to have higher availability, such as built-in hardware redundancy, hard disk, power, fan, and PCI hot swap and heat exchange, through RAID data protection, etc. When the hardware suddenly fails, the system or automatically switches to another device, or when the system is running, the system administrator replaces the device. Most importantly, the server running on the IA-64 architecture will be able to provide thermal replacement of memory and processor modules. In summary, new scalability, high-performance system I / O standards, modular, redundancy, and applicability can greatly improve system availability. High availability operation also requires powerful system management and monitoring systems, and can predict and detect IT infrastructure issues.
From manageability, one technology for management is the Intelligent Platform Management Interface, IPMI. It defines the ordinary interface of the intelligent hardware device, which is used to monitor server health, such as processors and memory errors, temperature, voltages, power supplies. This information can manage applications, web-based enterprise management criteria, when the web-based enterprise management criteria, the information can be diagnosed with hardware errors and recovery. In addition, the system management characteristics of the platform running on IA-64 ensure that managers obtain platform management information through multiple channels such as dial or LAN. This information can be obtained when the system does not respond when the operating system does not respond.
All in all, IA-64 is an ideal platform for electronic services / e-commerce and online environments. It can generate or encrypt and decrypt the keys extremely quickly. Similar to the situation where the PC world has, by standardizing a variety of applications and various operating systems, IA-64 will contribute to the convergence of the computer industry in a simple industry-standard hardware platform. When a customer configures their computing structure, standardized results will bring more choices and low costs.
For workstation
In order to deal with complex graphics, the workstation has to wait, the lengthy process is handled, and the slow screen refresh is especially annoyed. Today, the design of integrated circuits is more and more complex, and a single IC chip includes 10 million or more bytes - this is an amazing figure.
Even the most powerful workstation based on the 32-bit processing architecture does not effectively handle IC design issues, one of the biggest limits is the memory addressing capability. A 32-bit processor can directly address the maximum 4GB system memory, however, large-scale IC design will generate a file set of billion bytes. The work area also has a new problem. One must pay attention to the connection between the segments in a separate work section, and the two have increased work in a single work section. Complexity.
The IA-64 processor family can solve these bottlenecks through high performance, and EPIC will be able to better solve these problems by optimized applications, and more effectively better solve these problems, at the same time, support from hardware and software vendors Will confirm that the workstation solutions and systems will be more robust. It allows workstations to provide higher quality 3D effects, video audio information, and fast communication capabilities.
Future scenery unlimited
