Fujitsu PRIMEQUEST: standard platform with mainframe reliability
Critical applications have very high demands on the server platform. It should provide constant availability, minimal downtime, high performance and scalability. Traditionally, proprietary RISC / UNIX platforms and mainframes have been used for these purposes. Standard architecture servers did not have these qualities.
Now the situation has changed: in terms of RAS (Reliability, Availability, Serviceability - reliability, availability and ease of maintenance), some models of x86 servers have reached a whole new level, and the advanced vendors from RAS have implemented new generation Xeon E7 processors. characteristics of RAS characteristic of UNIX systems and mainframe hardware architectures. However, only a few players in the global server market have at the same time experience developing RISC architecture servers and mainframes, which allows them to create systems with the highest level of reliability and performance.
Fujitsu, a Japanese company that offers a new generation of second-generation PRIMEQUEST 2000 servers based on the most powerful Xeon E7 V3 processors, is an extremely reliable platform for data center infrastructures. PRIMEQUEST systems are already widely known for their record-breaking performance and fault tolerance. Features of the new generation of PRIMEQUEST servers include advanced RAS components with advanced self-healing features, including dynamic configuration changes and extended partitions, DDR4 memory resiliency, and a new family of Intel processors.
')
The models presented in May 2015 include the four-processor system PRIMEQUEST 2400E2, the eight-processor PRIMEQUEST 2800E2, and the eight-processor model PRIMEQUEST 2800B2. The transition from Xeon E7 v2 to Xeon E7 v3 processors allowed Fujitsu to increase significantly the performance / dollar ratio for PRIMEQUEST servers:
PRIMEQUEST platform is distinguished from other leading vendors ’business-critical x86 systems by Fujitsu’s unique experience embodied in its technology, which has been developing mainframes and powerful RISC / UNIX servers for more than a decade. Therefore, Fujitsu PRIMEQUEST 2000 Series servers are highly available UNIX-class and cost-effective standard architecture. They are even called x86 mainframes.
By the way, this company has already 50 years of experience in creating business-critical systems.
The first PRIMEQUEST system on the Intel x86 platform appeared in 2005. The newest Fujitsu PRIMEQUEST server line uses a highly reliable server architecture, many features of which are inherited from Fujitsu PRIMEPOWER UNIX servers and Fujitsu mainframes.
E2 models are designed for critical missions where downtime is unacceptable. These systems will be a good option for companies that wish to switch to the standard architecture platform without losing the reliability inherent in modern UNIX systems and mainframes.
The PRIMEQUEST 2800B2 model is a server for even more memory intensive applications such as in-memory calculations and SAP HANA. It provides cost effectiveness and guaranteed support for business-critical security levels.
Today, customers have a real alternative to expensive proprietary systems. An extensive ecosystem and attractive product prices have become weighty arguments in favor of x86 servers. It is not by chance that companies and organizations around the world are increasingly choosing cost-effective x86 servers for Linux and Windows, replacing their proprietary UNIX platforms: according to IDC, the number of such servers has grown by 25% -30% over the past five years. The emergence of new classes of tasks, including support for mobility, cloud computing, Big Data analysts, further increase the need for cost-effective business-critical servers.
As a standard architecture server, PRIMEQUEST gives customers access to the entire vast x86 ecosystem, allowing them to use a large number of applications created for this architecture. Such a platform can be used for the most demanding and resource-intensive workloads, including decision support systems, SAP and Oracle DBMS. Finally, a very important factor in modern conditions is price. PRIMEQUEST servers are characterized by the lowest price among enterprise-class systems at a level of reliability comparable to such UNIX platforms as IBM Power and HP Superdome 2.
PRIMEQUEST implements automated recovery of “motherboards” - server modules containing processors and RAM, flexible I / O redirection, maintenance without shutting down the system, redundant control processor and system clocks, memory mirroring, detection and correction of CPU errors, redundancy and hot replacement of components.
High requirements of applications to the level of accessibility of the hardware platform determine its features. It must ensure data integrity, failover recovery in minimal time and proactive management / monitoring to prevent downtime. At the same time, its architecture should be well balanced and scalable, not have “bottlenecks”.
The first PRIMEQUEST server, released in 2005, became the most powerful IA-64 server for critical computing at that time. In 2012, the PRIMEQUEST 1000 line was introduced, in 2014, deliveries of servers of the new series, PRIMEQUEST 2000, began. The last update dates from May 2015, when the E2 / B2 models were introduced.
Servers of this series were originally created as an x86 platform for running critical enterprise applications, but at the same time allowing to reduce the high costs traditionally inherent in the UNIX world, not inferior to UNIX systems in terms of availability, scalability and performance. Fujitsu PRIMEQUEST servers allow to solve problems of consolidation, standardization of the server park, to ensure the smooth operation of business-critical applications.
Servers support standard operating systems - RedHat and SuSE Linux or Windows Server. They can be used for large databases, in-memory databases such as SAP HANA, as well as for working with Big Data and as a platform for business-critical applications.
Since the requirements for high availability and cost-effectiveness of different customers vary considerably, Fujitsu offers three PRIMEQUEST models: 2800 / 2400E2 (Enterprise Model) with the highest level of availability and 2000B2 (Business Model), “honed” to achieve the best economic efficiency. Due to the high energy efficiency and reduced administration costs, these servers are one of the most economical systems of this class.
Fujitsu PRIMEQUEST 2400 / 2800E2 (figure 2 means the second generation of servers) - 4- / 8-socket system in the 10U package - positioned as a dynamic platform for a modern data center that combines the efficiency and flexibility of x86 servers with high availability and reliability of platforms SPARC.
Older PRIMEQUEST models support up to 8 Intel Xeon processors — up to 144 processing cores. Management Board Management (MMB) controls the operation of the server completely, collecting data and statistics from its components, monitoring the temperature inside the system, helping to identify and fix faults. For disks, processors and memory, potential problems are identified in advance, even before a possible failure.
MMB also controls the process of starting and stopping the server, activating and deactivating system resources. In addition, MMB in PRIMEQUEST uses the number of modules that are required in accordance with the power consumed at the moment, which makes a significant contribution to the energy efficiency of the system. Fan speed is also adjustable according to current cooling needs. Thus, MMB improves the reliability and energy efficiency of the system. MMB hardware controls are duplicated. For ease of administration, they are integrated with the ServerView Suite management software package.
PRIMEQUEST adjusts the number of active power supply units (PSUs) according to the server’s power consumption. Take, for example, the redundant configuration of the N + 1 power supply modules. If four PSU modules operate with a load of 40% of the maximum value, the server reduces the number of active PSU modules to three with a load of 53%. As a result, one PSU module becomes inactive.
All major components support hot swap: power modules, fans, disk drives, PCI cards, service processors. Almost all components are duplicated.
Thanks to hardware sections, the PRIMEQUEST system provides much higher reliability than platforms with software-only virtualization. It uses physical partitions (hardware partitions) and active redundancy of the so-called "System Board" (SB) - replaceable server modules.
Physical partitions are actually several independent systems with common I / O, operating on the same platform — an excellent opportunity for server consolidation.
Physical partitions can be reconfigured without rebooting the system. Dynamically reconfigurable partitions are a unique feature for x86 systems. For example, if the resources of one hardware partition are not enough to cope with the current load, you can allocate additional memory / processor power from another partition to it. Like this:
The partition configuration can be changed without shutting down the system and restarting it, adding resources to a specific partition or deleting them. This greatly increases the flexibility and reliability of the system.
You can reserve the System Board in case of a failure or to expand the working partition.
With increasing load, you can install another module (System Board) in a working system and add its resources to the appropriate section. In the same way, it is possible to change the equipment on a “live” system in case of failure of one “motherboard”.
Fujitsu offers a graphical interface for reconfiguring partitions. Working with him is not particularly difficult. It is quite intuitive.
However, command line lovers can use it. In this example, System Board 3 is added to section 0 live.
If the operation is successful, the green indicator on the front panel of the server will light up, indicating the operating status of the corresponding “motherboard”. Another example is removing System Board 2 from section 0:
Hardware partitions (PPAR) with dynamic reconfiguration at the level of "motherboards" is not limited. Fujitsu engineers have provided the so-called extended sections.
Extended partitions (EPAR) allow you to manipulate resources with greater granularity — not at the level of the entire System Board system module, but at the level of computational cores and memory blocks, which helps to optimize the use of system resources — memory and CPU, or isolate failures. In addition, selecting the required number of cores, you can significantly save on software licensing. For example, licenses for Oracle DBMS are very expensive.
The differences in the sections of different types are shown in the following table.
Thus, PRIMEQUEST servers support several types of partitions — physical, extended, and software (virtual machines) —with different levels of isolation of hardware and software errors and consolidation capabilities.
Summing up:
Sections:
PRIMEQUEST supports up to 8 electrically isolated partitions — hardware-level partitions, as well as extended partitions with functionality similar to virtual machines, but not requiring a hypervisor license.
Flexible I / O:
In PRIMEQUEST, the entire I / O subsystem is accessible to all CPUs through PCIe switches. Supports up to 56 PCI slots and the ability to install a large number of PCI SSD drives without losing I / O performance.
Hot Standby System Board:
In PRIMEQUEST, it can be used to recover from a motherboard failure, and dynamic reconfiguration is supported. Also, the backup system board can be added "live" to the partition. This is implemented for Windows and Linux.
Parallel service:
In PRIMEQUEST, individual sections can be stopped for maintenance, while other sections work. The control module is reserved.
As a result, PRIMEQUEST combines the qualities of standard x86 servers and the technologies of the PRIMEPOWER platform, ensuring high performance and availability.
According to Intel , thanks to Xeon E7 v3 processors, Fujitsu PRIMEQUEST servers increase the efficiency of IT infrastructure consolidation by 26% due to increased throughput when running applications and support 40% more virtual machines. Specifications of PRIMEQUEST 2800E2 and Xeon E7 8800 processor series can be found here .
With the transition to a new generation of Xeon v3, PRIMEQUEST's performance has grown by about 20%.
In May, the PRIMEQUEST 2800E2 servers with 8 Xeon E7-8890 v3 processors set a record among eight-processor systems when performing the SPECint_rate_base2006 benchmark, and the PRIMEQUEST 2800E2 server with 2 Xeon E7-8890 v3 benchmarks became the leader in the SPECfp_rate_base20066 test in the bench in the same mode in the past 20166 in the last year in the last run in the last day in the last year in the last year in the last year, and in the last place in the last year; The results of these and other tests can be found here .
When performing the SAP SD test, the Fujitsu PRIMEQUEST 2800E2 server served simultaneously over 56 thousand users.
The Model 2800E2 scales up to 8 Intel Xeon processors E7-8800 v3 processors and 12 TB of RAM (with Linux). It is characteristic that with the increase in the number of motherboards / processors, the server performance when working with a DBMS grows almost linearly, which indicates a balanced system architecture and its competent design. Especially high performance can be achieved when using an external flash array.
What is the use of such server platforms? They best reveal their qualities in such usage scenarios as mission-critical applications and scalable databases, real-time analytics, and server capacity consolidation. This platform actually combines up to eight physically independent servers, helping to significantly reduce operational costs.
PRIMEQUEST is focused on customers migrating applications from mainframes and RISC / UNIX systems , implementing large projects for scaling, virtualization and deployment of resource-intensive Linux / Windows applications, DBMS.
PRIMEQUEST allows you to minimize planned and unplanned downtime, since almost all system components are completely redundant and support hot swapping. In addition, CPUs and RAM are fault tolerant thanks to multi-level data protection mechanisms. And the performance of the PRIMEQUEST server scales very well to its maximum value.
Unique functionality and fault tolerance make this server a unique offer on the market, and the use of the latest components, such as Intel Xeon V3 and DDR4 memory, allows you to actively compete with servers that use 16 processors or more, but are based on less efficient components.
A complete set of error detection and correction tools, as well as high-availability system architecture, ensure the maximum uptime of the PRIMEQUEST server. This server can significantly reduce downtime due to maintenance operations, such as replacing failed components and updating firmware. In addition, due to the simple design, where most components are connected using a backplane almost without using cables, components can be replaced through the front or rear panel without removing equipment from the rack.
PRIMEQUEST servers are targeted at companies that need a highly reliable Linux-based or Windows Server-based system, organizations that have outgrown typical x86 servers, and those who would like to reduce operating costs and license fees.
Now the situation has changed: in terms of RAS (Reliability, Availability, Serviceability - reliability, availability and ease of maintenance), some models of x86 servers have reached a whole new level, and the advanced vendors from RAS have implemented new generation Xeon E7 processors. characteristics of RAS characteristic of UNIX systems and mainframe hardware architectures. However, only a few players in the global server market have at the same time experience developing RISC architecture servers and mainframes, which allows them to create systems with the highest level of reliability and performance.
Fujitsu, a Japanese company that offers a new generation of second-generation PRIMEQUEST 2000 servers based on the most powerful Xeon E7 V3 processors, is an extremely reliable platform for data center infrastructures. PRIMEQUEST systems are already widely known for their record-breaking performance and fault tolerance. Features of the new generation of PRIMEQUEST servers include advanced RAS components with advanced self-healing features, including dynamic configuration changes and extended partitions, DDR4 memory resiliency, and a new family of Intel processors.
')
The models presented in May 2015 include the four-processor system PRIMEQUEST 2400E2, the eight-processor PRIMEQUEST 2800E2, and the eight-processor model PRIMEQUEST 2800B2. The transition from Xeon E7 v2 to Xeon E7 v3 processors allowed Fujitsu to increase significantly the performance / dollar ratio for PRIMEQUEST servers:
PRIMEQUEST platform is distinguished from other leading vendors ’business-critical x86 systems by Fujitsu’s unique experience embodied in its technology, which has been developing mainframes and powerful RISC / UNIX servers for more than a decade. Therefore, Fujitsu PRIMEQUEST 2000 Series servers are highly available UNIX-class and cost-effective standard architecture. They are even called x86 mainframes.
By the way, this company has already 50 years of experience in creating business-critical systems.
The first PRIMEQUEST system on the Intel x86 platform appeared in 2005. The newest Fujitsu PRIMEQUEST server line uses a highly reliable server architecture, many features of which are inherited from Fujitsu PRIMEPOWER UNIX servers and Fujitsu mainframes.
E2 models are designed for critical missions where downtime is unacceptable. These systems will be a good option for companies that wish to switch to the standard architecture platform without losing the reliability inherent in modern UNIX systems and mainframes.
The PRIMEQUEST 2800B2 model is a server for even more memory intensive applications such as in-memory calculations and SAP HANA. It provides cost effectiveness and guaranteed support for business-critical security levels.
Why x86?
Today, customers have a real alternative to expensive proprietary systems. An extensive ecosystem and attractive product prices have become weighty arguments in favor of x86 servers. It is not by chance that companies and organizations around the world are increasingly choosing cost-effective x86 servers for Linux and Windows, replacing their proprietary UNIX platforms: according to IDC, the number of such servers has grown by 25% -30% over the past five years. The emergence of new classes of tasks, including support for mobility, cloud computing, Big Data analysts, further increase the need for cost-effective business-critical servers.
As a standard architecture server, PRIMEQUEST gives customers access to the entire vast x86 ecosystem, allowing them to use a large number of applications created for this architecture. Such a platform can be used for the most demanding and resource-intensive workloads, including decision support systems, SAP and Oracle DBMS. Finally, a very important factor in modern conditions is price. PRIMEQUEST servers are characterized by the lowest price among enterprise-class systems at a level of reliability comparable to such UNIX platforms as IBM Power and HP Superdome 2.
PRIMEQUEST implements automated recovery of “motherboards” - server modules containing processors and RAM, flexible I / O redirection, maintenance without shutting down the system, redundant control processor and system clocks, memory mirroring, detection and correction of CPU errors, redundancy and hot replacement of components.
High requirements of applications to the level of accessibility of the hardware platform determine its features. It must ensure data integrity, failover recovery in minimal time and proactive management / monitoring to prevent downtime. At the same time, its architecture should be well balanced and scalable, not have “bottlenecks”.
PRIMEQUEST architecture
The first PRIMEQUEST server, released in 2005, became the most powerful IA-64 server for critical computing at that time. In 2012, the PRIMEQUEST 1000 line was introduced, in 2014, deliveries of servers of the new series, PRIMEQUEST 2000, began. The last update dates from May 2015, when the E2 / B2 models were introduced.
Servers of this series were originally created as an x86 platform for running critical enterprise applications, but at the same time allowing to reduce the high costs traditionally inherent in the UNIX world, not inferior to UNIX systems in terms of availability, scalability and performance. Fujitsu PRIMEQUEST servers allow to solve problems of consolidation, standardization of the server park, to ensure the smooth operation of business-critical applications.
Servers support standard operating systems - RedHat and SuSE Linux or Windows Server. They can be used for large databases, in-memory databases such as SAP HANA, as well as for working with Big Data and as a platform for business-critical applications.
Since the requirements for high availability and cost-effectiveness of different customers vary considerably, Fujitsu offers three PRIMEQUEST models: 2800 / 2400E2 (Enterprise Model) with the highest level of availability and 2000B2 (Business Model), “honed” to achieve the best economic efficiency. Due to the high energy efficiency and reduced administration costs, these servers are one of the most economical systems of this class.
Fujitsu PRIMEQUEST 2400 / 2800E2 (figure 2 means the second generation of servers) - 4- / 8-socket system in the 10U package - positioned as a dynamic platform for a modern data center that combines the efficiency and flexibility of x86 servers with high availability and reliability of platforms SPARC.
Older PRIMEQUEST models support up to 8 Intel Xeon processors — up to 144 processing cores. Management Board Management (MMB) controls the operation of the server completely, collecting data and statistics from its components, monitoring the temperature inside the system, helping to identify and fix faults. For disks, processors and memory, potential problems are identified in advance, even before a possible failure.
MMB also controls the process of starting and stopping the server, activating and deactivating system resources. In addition, MMB in PRIMEQUEST uses the number of modules that are required in accordance with the power consumed at the moment, which makes a significant contribution to the energy efficiency of the system. Fan speed is also adjustable according to current cooling needs. Thus, MMB improves the reliability and energy efficiency of the system. MMB hardware controls are duplicated. For ease of administration, they are integrated with the ServerView Suite management software package.
PRIMEQUEST adjusts the number of active power supply units (PSUs) according to the server’s power consumption. Take, for example, the redundant configuration of the N + 1 power supply modules. If four PSU modules operate with a load of 40% of the maximum value, the server reduces the number of active PSU modules to three with a load of 53%. As a result, one PSU module becomes inactive.
All major components support hot swap: power modules, fans, disk drives, PCI cards, service processors. Almost all components are duplicated.
Partitions and dynamic reconfiguration
Thanks to hardware sections, the PRIMEQUEST system provides much higher reliability than platforms with software-only virtualization. It uses physical partitions (hardware partitions) and active redundancy of the so-called "System Board" (SB) - replaceable server modules.
Physical partitions are actually several independent systems with common I / O, operating on the same platform — an excellent opportunity for server consolidation.
Physical partitions can be reconfigured without rebooting the system. Dynamically reconfigurable partitions are a unique feature for x86 systems. For example, if the resources of one hardware partition are not enough to cope with the current load, you can allocate additional memory / processor power from another partition to it. Like this:
The partition configuration can be changed without shutting down the system and restarting it, adding resources to a specific partition or deleting them. This greatly increases the flexibility and reliability of the system.
You can reserve the System Board in case of a failure or to expand the working partition.
With increasing load, you can install another module (System Board) in a working system and add its resources to the appropriate section. In the same way, it is possible to change the equipment on a “live” system in case of failure of one “motherboard”.
Fujitsu offers a graphical interface for reconfiguring partitions. Working with him is not particularly difficult. It is quite intuitive.
However, command line lovers can use it. In this example, System Board 3 is added to section 0 live.
If the operation is successful, the green indicator on the front panel of the server will light up, indicating the operating status of the corresponding “motherboard”. Another example is removing System Board 2 from section 0:
Hardware partitions (PPAR) with dynamic reconfiguration at the level of "motherboards" is not limited. Fujitsu engineers have provided the so-called extended sections.
Extended partitions (EPAR) allow you to manipulate resources with greater granularity — not at the level of the entire System Board system module, but at the level of computational cores and memory blocks, which helps to optimize the use of system resources — memory and CPU, or isolate failures. In addition, selecting the required number of cores, you can significantly save on software licensing. For example, licenses for Oracle DBMS are very expensive.
The differences in the sections of different types are shown in the following table.
Thus, PRIMEQUEST servers support several types of partitions — physical, extended, and software (virtual machines) —with different levels of isolation of hardware and software errors and consolidation capabilities.
Summing up:
Sections:
PRIMEQUEST supports up to 8 electrically isolated partitions — hardware-level partitions, as well as extended partitions with functionality similar to virtual machines, but not requiring a hypervisor license.
Flexible I / O:
In PRIMEQUEST, the entire I / O subsystem is accessible to all CPUs through PCIe switches. Supports up to 56 PCI slots and the ability to install a large number of PCI SSD drives without losing I / O performance.
Hot Standby System Board:
In PRIMEQUEST, it can be used to recover from a motherboard failure, and dynamic reconfiguration is supported. Also, the backup system board can be added "live" to the partition. This is implemented for Windows and Linux.
Parallel service:
In PRIMEQUEST, individual sections can be stopped for maintenance, while other sections work. The control module is reserved.
As a result, PRIMEQUEST combines the qualities of standard x86 servers and the technologies of the PRIMEPOWER platform, ensuring high performance and availability.
Performance
According to Intel , thanks to Xeon E7 v3 processors, Fujitsu PRIMEQUEST servers increase the efficiency of IT infrastructure consolidation by 26% due to increased throughput when running applications and support 40% more virtual machines. Specifications of PRIMEQUEST 2800E2 and Xeon E7 8800 processor series can be found here .
With the transition to a new generation of Xeon v3, PRIMEQUEST's performance has grown by about 20%.
In May, the PRIMEQUEST 2800E2 servers with 8 Xeon E7-8890 v3 processors set a record among eight-processor systems when performing the SPECint_rate_base2006 benchmark, and the PRIMEQUEST 2800E2 server with 2 Xeon E7-8890 v3 benchmarks became the leader in the SPECfp_rate_base20066 test in the bench in the same mode in the past 20166 in the last year in the last run in the last day in the last year in the last year in the last year, and in the last place in the last year; The results of these and other tests can be found here .
When performing the SAP SD test, the Fujitsu PRIMEQUEST 2800E2 server served simultaneously over 56 thousand users.
The Model 2800E2 scales up to 8 Intel Xeon processors E7-8800 v3 processors and 12 TB of RAM (with Linux). It is characteristic that with the increase in the number of motherboards / processors, the server performance when working with a DBMS grows almost linearly, which indicates a balanced system architecture and its competent design. Especially high performance can be achieved when using an external flash array.
Summary
What is the use of such server platforms? They best reveal their qualities in such usage scenarios as mission-critical applications and scalable databases, real-time analytics, and server capacity consolidation. This platform actually combines up to eight physically independent servers, helping to significantly reduce operational costs.
PRIMEQUEST is focused on customers migrating applications from mainframes and RISC / UNIX systems , implementing large projects for scaling, virtualization and deployment of resource-intensive Linux / Windows applications, DBMS.
PRIMEQUEST allows you to minimize planned and unplanned downtime, since almost all system components are completely redundant and support hot swapping. In addition, CPUs and RAM are fault tolerant thanks to multi-level data protection mechanisms. And the performance of the PRIMEQUEST server scales very well to its maximum value.
Unique functionality and fault tolerance make this server a unique offer on the market, and the use of the latest components, such as Intel Xeon V3 and DDR4 memory, allows you to actively compete with servers that use 16 processors or more, but are based on less efficient components.
A complete set of error detection and correction tools, as well as high-availability system architecture, ensure the maximum uptime of the PRIMEQUEST server. This server can significantly reduce downtime due to maintenance operations, such as replacing failed components and updating firmware. In addition, due to the simple design, where most components are connected using a backplane almost without using cables, components can be replaced through the front or rear panel without removing equipment from the rack.
PRIMEQUEST servers are targeted at companies that need a highly reliable Linux-based or Windows Server-based system, organizations that have outgrown typical x86 servers, and those who would like to reduce operating costs and license fees.
Source: https://habr.com/ru/post/263779/
All Articles