HP Integrity NonStop: Availability Is Key
The main advantages of the HP NonStop MPP platform are high availability and scalability that are not achievable for SMP systems.
The topic of parallel computing platforms such as HP NonStop is rarely covered in the press, due to the fact that these platforms are specialized: in the world, 80-90% of these systems are used in the banking sector (card processing), and the rest are mainly in telecom ( HLR and billing). One-time implementations take place as railroad traffic control systems, production processes, support for postal services, exchange trading, ticket booking, etc. As a rule, applications for these tasks are not commercial products, but are developed by customers for their own needs. In Russia, HP NonStop systems today are used only by banks and telecommunications companies.
Back in March 2014, the company introduced the new NonStop systems with Intel Itanium processors with higher performance than previous systems. And in March 2015, HP introduced to the market fail-safe NonStop systems based on x86 high end architecture. These new HP systems fit well with the market trend — using servers on increasingly powerful Intel Xeon processors for mission-critical business tasks.
')
New HP NonStop X series servers are available for order from mid-March 2015, including in Russia.
NonStop X systems contain up to 16 Intel Xeon processors and up to 3 TB of RAM. Interconnect is InfiniBand. These systems can be clustered with 4080 processors. In the future, the corporation plans to supplement the NonStop X line with entry-level systems and specialized servers for telecom operators.
Although the main attention in the new systems is attracted by the hardware, the main thing is in the system software. Despite the emergence of the NonStop X model on the x86 platform, NonStop is not at all a competitor to the older versions of the x86 servers of the same HP - these are solutions of a fundamentally different class. Unlike the Superdome, HP NonStop is still not a platform for either Linux or Windows. As well as throughout its evolution, in terms of hardware and software architecture, this is all the same NonStop platform with all its inherent advantages.
HP Integrity NonStop, as before, uses a specialized NonStop OS Kernel and NonStop SQL database and is a deeply integrated hardware and software system. HP has been constantly working to unify the hardware components of the system, starting with storage subsystems and memory components, ending with processors and using a blade architecture (c7000 chassis). In fact, until the advent of NonStop on x86, the ServerNet system switch remained the only proprietary component, performing message routing, which can be considered the prototype of InfiniBand. As already noted, in the new version of NonStop on x86, InfiniBand came to replace ServerNet.
The appearance of the non86 model on the x86 platform is primarily due to the company's commitment to the strategy of unifying the component base, which ultimately affects the cost of the systems and provides additional benefits for HP NonStop users. In addition, with the use of Intel Xeon processors, it was possible to improve the performance of the platform as a whole, which in turn makes HP NonStop X more satisfying to the modern needs of users of the HP NonStop platform.
However, using x86 did not turn NonStop into a standard architecture system. The reason is that it is the system software that largely determines the highest reliability of this platform and its scalability. Linear scalability availability of applications at the level of "five nines" - the main advantages of NonStop - remain unshakable on x86.
In 2011, HP announced the Odyssey project, aimed at creating a system based on Xeon, suitable for the most demanding loads. Superdome X and NonStop X represent the most important results of these efforts.
Blade architecture in the service of NonStop. The c7000 chassis is used by HP in the ProLiant, Superdome and Integrity NonStop server lines. NonStop blades are single-processor (logical processor). This design improves serviceability, and differs from IB interconnect from the usual c7000 chassis.
So, the NonStop X system software has remained the same and has changed only in terms of the capabilities of working with Intel Xeon. As before, each HP NonStop server is a system of massively parallel architecture that does not contain shared resources and forms a loosely coupled cluster, the objects of which are not only processors, but also input / output controllers and disk subsystems. All server objects are connected via duplicated ServerNet or Infiniband networks (depending on the type of system) that replace the server's system bus.
NonStop servers are combined into a cluster system based on the communication technologies described above, as well as using a special technology based on IP connections. The uniqueness of the HP NonStop architecture is its complete decentralization - all cluster objects are equal and duplicated. All hard drives are mirrored.
The logical processor has its own command device, RAM, memory access channels and connection routers. A separate copy of the operating system is executed on each such processor, and when the next processor is added to the system, its performance is fully adapted to the needs of the application tasks.
This distribution of the operating system erases the physical boundaries between the processors, allowing each operation performed on any of them to have access to any other system resources. Working independently of each other, the processors nevertheless act in concert thanks to the interprocessor messaging system.
Continuous availability of the server is ensured by using special technological schemes that isolate software applications from failures in the hardware and operating system. Continuous availability means that recovery from hardware or software failures is immediate and transparent to executable applications and users.
Separate consideration is deserved by the unique technology of pair processes. As mentioned earlier, each server processor runs its own copy of the operating system, which allows you to run different software processes on different processors. At the same time, each process running on any processor has its own backup copy on the other, where it is inactive and occupies almost no CPU time.
The primary process at certain times (usually before / after an I / O procedure or after completion of a transaction) sends control messages to the backup process containing the full content of the primary process. If, for some reason, the primary process terminates abnormally, then the backup process takes control and continues to run on another processor since the last control message.
We are talking here about the continuation of the program process (take over), and not about restarting it (restart). The computing power of NonStop servers is linearly scaled to 4080 processors due to the use of a massively parallel architecture with a weak connection between processors without resource sharing and allows you to connect all processors among themselves, as well as with input / output devices. This architecture prevents bottlenecks due to conflicts when sharing resources.
In addition to high RAS (Reliability, Availability, Serviceability), the NonStop system is highly scalable. In contrast to systems with a symmetric SMP architecture, the NonStop platform is a system with a mass-parallel architecture (MPP). There are no so-called shared resources in it. For example, processors use not common, but own allocated memory, which other processors do not work with. The CPU + memory with the “strapping” in the form of controllers and other auxiliary components of the system is a logical processor in NonStop terminology. The entire NonStop architecture ultimately resembles a compute cluster.
Since there is no system bus in NonStop, processors do not compete for access to it. The system bus in NonStop replaces the message router that the system components exchange (message-based OS), and not necessarily with the participation of the processor. As a result, it is possible to avoid the large number of collisions inherent in the traditional multiprocessor system when working with the system bus and memory, as well as resulting from this performance degradation with an increase in the number of processors and cores.
NonStop performance grows linearly, which is unattainable for SMP architecture platforms. Why is it important? Scalability allows the customer to increase the acquired system almost unlimitedly, and not to acquire a new system when the existing ceiling of the ability to upgrade is achieved, which negatively affects the cost of ownership (TCO). It also provides investment protection: NonStop systems of different generations can operate in the same cluster.
Currently, the NonStop line includes four server models: HP Integrity NonStop X NS7 X1 on Intel Xeon E5-2600 v2, HP Integrity NonStop NS2300, HP Integrity NonStop NS2400 and HP Integrity NonStop BladeSystem NB56000c. NonStop servers use Intel Itanium 9500 series processors.
There are no signs that Itanium will leave the market. So, when starting to use x86 processors in NonStop servers, the corporation will adhere to the Itanium-based systems release plan in accordance with the previously announced RoadMap.
If the scale of tasks and load is constantly increasing, a gradual increase in computing power is required, if the availability of a computing platform is critical for the image of a business or leads to large losses, NonStop systems can be the optimal choice. They do an excellent job with uneven transactional load (OLTP), well-parallelizable tasks, and can play the role of a database server. Systems are promising in a number of market segments, but the problem may be to find qualified developers who are familiar with this architecture.
Our previous publications:
»The correct bonus for your new server
» Cost-effective HP servers for SMB and providers
» HP OpenStack Helion - First Meet
» SDN: alternative or addition to traditional networks?
» New network architectures: open or closed solutions?
» Implementing MSA in a virtualized enterprise environment
» HP MSA Disk Arrays as a Basis for Data Consolidation
» Multivendor corporate network: myths and reality
» Available HP ProLiant server models (10 and 100 series)
» Convergence based on HP Networking. Part 1
» HP ProLiant ML350 Gen9 - server with insane extensibility
Thanks for attention!
The topic of parallel computing platforms such as HP NonStop is rarely covered in the press, due to the fact that these platforms are specialized: in the world, 80-90% of these systems are used in the banking sector (card processing), and the rest are mainly in telecom ( HLR and billing). One-time implementations take place as railroad traffic control systems, production processes, support for postal services, exchange trading, ticket booking, etc. As a rule, applications for these tasks are not commercial products, but are developed by customers for their own needs. In Russia, HP NonStop systems today are used only by banks and telecommunications companies.
NonStop on x86: how to understand this?
Back in March 2014, the company introduced the new NonStop systems with Intel Itanium processors with higher performance than previous systems. And in March 2015, HP introduced to the market fail-safe NonStop systems based on x86 high end architecture. These new HP systems fit well with the market trend — using servers on increasingly powerful Intel Xeon processors for mission-critical business tasks.
')
New HP NonStop X series servers are available for order from mid-March 2015, including in Russia.
NonStop X systems contain up to 16 Intel Xeon processors and up to 3 TB of RAM. Interconnect is InfiniBand. These systems can be clustered with 4080 processors. In the future, the corporation plans to supplement the NonStop X line with entry-level systems and specialized servers for telecom operators.
Although the main attention in the new systems is attracted by the hardware, the main thing is in the system software. Despite the emergence of the NonStop X model on the x86 platform, NonStop is not at all a competitor to the older versions of the x86 servers of the same HP - these are solutions of a fundamentally different class. Unlike the Superdome, HP NonStop is still not a platform for either Linux or Windows. As well as throughout its evolution, in terms of hardware and software architecture, this is all the same NonStop platform with all its inherent advantages.
HP Integrity NonStop, as before, uses a specialized NonStop OS Kernel and NonStop SQL database and is a deeply integrated hardware and software system. HP has been constantly working to unify the hardware components of the system, starting with storage subsystems and memory components, ending with processors and using a blade architecture (c7000 chassis). In fact, until the advent of NonStop on x86, the ServerNet system switch remained the only proprietary component, performing message routing, which can be considered the prototype of InfiniBand. As already noted, in the new version of NonStop on x86, InfiniBand came to replace ServerNet.
The appearance of the non86 model on the x86 platform is primarily due to the company's commitment to the strategy of unifying the component base, which ultimately affects the cost of the systems and provides additional benefits for HP NonStop users. In addition, with the use of Intel Xeon processors, it was possible to improve the performance of the platform as a whole, which in turn makes HP NonStop X more satisfying to the modern needs of users of the HP NonStop platform.
However, using x86 did not turn NonStop into a standard architecture system. The reason is that it is the system software that largely determines the highest reliability of this platform and its scalability. Linear scalability availability of applications at the level of "five nines" - the main advantages of NonStop - remain unshakable on x86.
In 2011, HP announced the Odyssey project, aimed at creating a system based on Xeon, suitable for the most demanding loads. Superdome X and NonStop X represent the most important results of these efforts.
Blade architecture in the service of NonStop. The c7000 chassis is used by HP in the ProLiant, Superdome and Integrity NonStop server lines. NonStop blades are single-processor (logical processor). This design improves serviceability, and differs from IB interconnect from the usual c7000 chassis.
New Platform - Old Architecture
So, the NonStop X system software has remained the same and has changed only in terms of the capabilities of working with Intel Xeon. As before, each HP NonStop server is a system of massively parallel architecture that does not contain shared resources and forms a loosely coupled cluster, the objects of which are not only processors, but also input / output controllers and disk subsystems. All server objects are connected via duplicated ServerNet or Infiniband networks (depending on the type of system) that replace the server's system bus.
NonStop servers are combined into a cluster system based on the communication technologies described above, as well as using a special technology based on IP connections. The uniqueness of the HP NonStop architecture is its complete decentralization - all cluster objects are equal and duplicated. All hard drives are mirrored.
The logical processor has its own command device, RAM, memory access channels and connection routers. A separate copy of the operating system is executed on each such processor, and when the next processor is added to the system, its performance is fully adapted to the needs of the application tasks.
This distribution of the operating system erases the physical boundaries between the processors, allowing each operation performed on any of them to have access to any other system resources. Working independently of each other, the processors nevertheless act in concert thanks to the interprocessor messaging system.
Continuous availability of the server is ensured by using special technological schemes that isolate software applications from failures in the hardware and operating system. Continuous availability means that recovery from hardware or software failures is immediate and transparent to executable applications and users.
Separate consideration is deserved by the unique technology of pair processes. As mentioned earlier, each server processor runs its own copy of the operating system, which allows you to run different software processes on different processors. At the same time, each process running on any processor has its own backup copy on the other, where it is inactive and occupies almost no CPU time.
The primary process at certain times (usually before / after an I / O procedure or after completion of a transaction) sends control messages to the backup process containing the full content of the primary process. If, for some reason, the primary process terminates abnormally, then the backup process takes control and continues to run on another processor since the last control message.
We are talking here about the continuation of the program process (take over), and not about restarting it (restart). The computing power of NonStop servers is linearly scaled to 4080 processors due to the use of a massively parallel architecture with a weak connection between processors without resource sharing and allows you to connect all processors among themselves, as well as with input / output devices. This architecture prevents bottlenecks due to conflicts when sharing resources.
Scalability
In addition to high RAS (Reliability, Availability, Serviceability), the NonStop system is highly scalable. In contrast to systems with a symmetric SMP architecture, the NonStop platform is a system with a mass-parallel architecture (MPP). There are no so-called shared resources in it. For example, processors use not common, but own allocated memory, which other processors do not work with. The CPU + memory with the “strapping” in the form of controllers and other auxiliary components of the system is a logical processor in NonStop terminology. The entire NonStop architecture ultimately resembles a compute cluster.
Since there is no system bus in NonStop, processors do not compete for access to it. The system bus in NonStop replaces the message router that the system components exchange (message-based OS), and not necessarily with the participation of the processor. As a result, it is possible to avoid the large number of collisions inherent in the traditional multiprocessor system when working with the system bus and memory, as well as resulting from this performance degradation with an increase in the number of processors and cores.
NonStop performance grows linearly, which is unattainable for SMP architecture platforms. Why is it important? Scalability allows the customer to increase the acquired system almost unlimitedly, and not to acquire a new system when the existing ceiling of the ability to upgrade is achieved, which negatively affects the cost of ownership (TCO). It also provides investment protection: NonStop systems of different generations can operate in the same cluster.
Future NonStop
Currently, the NonStop line includes four server models: HP Integrity NonStop X NS7 X1 on Intel Xeon E5-2600 v2, HP Integrity NonStop NS2300, HP Integrity NonStop NS2400 and HP Integrity NonStop BladeSystem NB56000c. NonStop servers use Intel Itanium 9500 series processors.
There are no signs that Itanium will leave the market. So, when starting to use x86 processors in NonStop servers, the corporation will adhere to the Itanium-based systems release plan in accordance with the previously announced RoadMap.
findings
If the scale of tasks and load is constantly increasing, a gradual increase in computing power is required, if the availability of a computing platform is critical for the image of a business or leads to large losses, NonStop systems can be the optimal choice. They do an excellent job with uneven transactional load (OLTP), well-parallelizable tasks, and can play the role of a database server. Systems are promising in a number of market segments, but the problem may be to find qualified developers who are familiar with this architecture.
Our previous publications:
»The correct bonus for your new server
» Cost-effective HP servers for SMB and providers
» HP OpenStack Helion - First Meet
» SDN: alternative or addition to traditional networks?
» New network architectures: open or closed solutions?
» Implementing MSA in a virtualized enterprise environment
» HP MSA Disk Arrays as a Basis for Data Consolidation
» Multivendor corporate network: myths and reality
» Available HP ProLiant server models (10 and 100 series)
» Convergence based on HP Networking. Part 1
» HP ProLiant ML350 Gen9 - server with insane extensibility
Thanks for attention!
Source: https://habr.com/ru/post/256177/
All Articles