From “Business- to Mission critical”: Fujitsu server overview
- It is not surprising that he covered himself. It says "Made in Japan"
- What are you talking about, Doc? All the best gizmos are made in Japan.
- Incredible ...
x / f "Back to the Future"
No less emotionally than the presidency of film actor Ronald Reagan in the “far” 1985, Dr. Emmett Latrop von Braun from 1955 was struck with a remark from the 1985 “boy from the future” Marty McFly that the best electronics in the “far future” produced in Japan. The fact is that before the “economic miracle” of the 70s of the last century, the goods produced in Japan had a low-quality reputation because of the industry destroyed during the Second World War. By the way, a similar stamp of “low-grade junk” in the post-war period was worn in goods made in Germany for exactly the same reasons. And the “engineer Kaiser” was still to put an end to this negligence and to go all the way back to the chased phrase “German quality,” which the Germans did with success.
Presented in the film “Back to the Future-2” by R. Zemeckis, the world of 2015, of course, is different from what we see now behind the window. There were more cars, but they did not take off, but got stuck in traffic jams. Skateboards are still on wheels, although Lexus is trying to “hang” the board with liquid nitrogen. But, probably, if from our 2015, go back to the slightly tuned DeLorean DMC-12, like Marty McFly, 30 years ago, in his native 1985, and show him a broken iPhone 6, Marty will speak out similarly to Doc: “Not surprising that he covered himself. It says, “Made in China.” ”
')
The dominant position of globalization and the struggle of producers for reducing production and logistics costs in the past 30 years have taught us that far from the cheapest products under respectable brands are produced and assembled in countries whose attitude to production capabilities among the masses once was, let's say skeptical. For example, telecommunications and server equipment, as a rule, are produced and assembled in Taiwan, China, India, Mexico, Ireland, Poland, Bulgaria, and the Czech Republic. But despite the domination of globalization and the dictates of cost reduction, there are still manufacturers in our world, whose products contain special and capacious words for the older generation. In this article, we’ll tell you about Fujitsu servers that carry the proud “old school” stamp: “Made in Japan” or “Made in Germany”.
Separately, we note that in view of the geopolitical factor, American manufacturers of IT equipment have stopped selling products to a number of government organizations and companies with state participation, and more recently, to some private companies. The general public is aware that the Chinese manufacturers have been very enthusiastic about these circumstances and wish them good health as long as possible. But it will not be superfluous to know that being under Japanese jurisdiction and, what is more important in this case, based on the strong national support and domestic demand of Fujitsu Limited of the Japanese market, Russian so-called “sanctions” customers are quite loyal.
The Japanese company Fujitsu has been manufacturing standard architecture (x86) servers for 20 years. Probably, it can be said that the fact that Fujitsu’s undisputed championship in the national market for server solutions is not the most convincing due to current ideas about the specifics and mentality of the Japanese with a focus on products of national manufacturers, but still it is first in the third largest (in terms of GDP) the economy of the world after the USA and China. In the world market, the company ranks 4th. In other words, Fujitsu is no stranger to the server solutions market and has extensive experience.
The portfolio of Fujitsu standard server solutions (x86) consists of two lines: PRIMERGY and PRIMEQUEST. The PRIMERGY line includes standard rack and tower (floor) servers, as well as Blade servers and CX servers for high performance computing (HPC) and horizontal scaling. The PRIMEQUEST family of servers is focused on high-performance devices based on Intel Xeon E7 with increased component reliability and hardware fault tolerance.
PRIMERGY servers are manufactured at the Fujitsu plant in Augsburg, Germany, while PRIMEQUEST servers are built in Japan, although the possibility of expanding the plant in Bavarian Augsburg and launching an additional PRIMEQUEST assembly line is being considered. Touching on the topic of production, it is worth noting that Fujitsu declares the availability of components to its products for an average of five years from the moment of the official withdrawal of the model from the sale; this is the standard manufacturer's policy. At the same time, the customer has the opportunity to extend the availability of spare parts using a special contract, which will be reserved for the customer. Such an option can be an alternative for those who, due to certain reasons (type of tasks solved by the server, server location, economic reasons), intend to use the server for a sufficiently long period, but do not have the need or the ability to pre-purchase from the warehouse (and for years to withdraw the necessary for storage warehouse space) a list of server spare parts.
Objectively comparing servers of different manufacturers in terms of price / performance ratio, especially in Russia, is quite a difficult task, in the Russian Federation this is complicated by the pricing system used by manufacturers and the sales model, even the exchange selling price for goods may differ depending on the project conditions. Anyone who has had to deal with a marketing presentation of equipment manufacturers more than once knows that all manufacturers “have up to date” any report or review from a specialized organization that shows leadership in various parameters of the manufacturer’s products against direct competitors. As mentioned above, Fujitsu is not the first day on the market and, accordingly, has received such positive reports for itself - for example, from the Transaction Processing Performance Council (TPC), registered in 1988, which specializes in evaluating the performance of computing systems in processing transactions and work with databases. The TPC-E test report is available at the link ; we leave you to assess the credibility of the material.
In the line, as mentioned above, the servers are rack and tower versions. The model number should be read as follows: the letter R, T or B indicates the server form factor, respectively, rack (R - rack), tower (T - tower) or blade server (B - blade). Next comes a slight (temporary) confusion. Articles of the previous generation of Fujitsu PRIMERGY servers consisted of a three-digit numeric number and an “S” at the end. The current generation has a four-digit numerical number in the article and “M1” at the end. The three-digit article of the servers of the previous generation today can only tell the user that this is the previous generation and the server is obsolete, whereas in the new generation the manufacturer has made the articles more informative. The first digit of the number indicates the possible number of processors (socket) in the server. The second digit is the series number of the used Intel Xeon processors, i.e. E3, E5, E7 (respectively, "3", "5", "7"). For example: RX4770 M1 is a rack server with four Intel Xeon E7 sockets, TX2540 M1 is a tower server with two Intel Xeon E5 sockets. In our opinion, the manufacturer offered a fairly convenient model for marking devices, which allows without going into the specification to “read” some of the basic characteristics of the server according to the article, and sometimes avoid extremely annoying (and expensive) errors.
A number of PRIMERGY server models (mostly older models) support, in their series, technologies for the developer and manufacturer of network solutions from Emulex, including Emulex Universal Multi-Channel technology (UMC), which allows you to create multiple PCI functions on each network adapter. For example, each 10GbE port can be configured with 8 functions of the virtual network adapter. The user has the ability to configure the bandwidth of each channel and traffic separation, the entire system will work correctly with any 10GbE switch.
As already noted, the Japanese manufacturer in the design of PRIMERGY servers paid great attention to the architecture and layout of the servers. The servers are not only energy efficient (unfortunately, we often don’t care about this customer’s indicator) and are quiet (this indicator is much more interesting), but they can also work 24/7 at temperatures up to 40 ° C.
Fujitsu PRIMERGY RX Rack Servers
At present, the PRIMERGY server rack group of the new generation includes: single-unit servers RX1330 M1 (with one Intel Xeon E3 processor) and RX2530 M1 (up to two processors). Two dual-module dual-processor models RX2520 M1 and RX2540 M1, two four-unit models RX2560 M1 and RX4770 M1 (with the possibility of installing up to 4 processors). Also currently available to order servers from the previous generation: dual-dual dual-processor RX300 S8 and four-dual dual-processor RX350 S8.
Fujitsu PRIMERGY RX2520 M1
The differences between the servers are standard: the number of installed processors and the power consumed by them, the number of slots and the maximum amount of RAM, disks, the number of installed PCIe cards, etc.
In addition to the two low-end models, the RX1330 M1 and RX2520 M1, all Fujitsu PRIMERGY rack-mount servers, including the single-unit RX2530 M1, support expandable RAM up to 1536 GB or more (in 4U models). The servers use only disks of large (LFF, 3.5 ″) and small (SFF, 2.5 ″) form factors, the manufacturer uses 1.8 ″ disks only in blade servers. Some models of Fujitsu PRIMERGY rack servers have the ability to install PCIe-SSD disks, which are becoming more common for solving problems that require fast SSD disks for caching (for example, in the implementation of the so-called Software Defined solutions).
In its servers (high models), Fujitsu offers an option that is not often seen from other manufacturers - the ability to install internal DOM (Disk on Module) drives with SATA 3G interface into the SATA connector directly on the motherboard. Rather inexpensive and energy-efficient drives, with reliability significantly higher than that of a traditional USB flash drive or SD card. Such drives are designed primarily to perform the functions of a boot disk and are an alternative solution in a common scheme with SD cards (single or duplicated).
Also, older models support DynamicLoM technology. This solution, which essentially pre-installs a network adapter at a speed of up to 40 Gbps onboard the server's motherboard, while the user chooses the way out based on his needs. That is, you can use 1 Gbps output and modify it to 10 or 40 Gbps (in the near future), depending on the needs.
Fujitsu PRIMERGY TX Tower Servers
The PRIMERGY tower server group consists of three single-processor servers TX1310 M1, TX1320 M1, TX1330 M1 and two dual-processor servers TX2540 M1, TX2560 M1. From the previous generation, the single-processor TX150 S8 and the dual-processor TX300 S8 are still available for order.
Fujitsu PRIMERGY TX2560 M1
Fujitsu TX tower servers, depending on the model, can work with 1 or 2 Intel Xeon E3 and E5 processors. Uses large (LFF, 3.5 ") and small (SFF, 2.5") form factors. In all servers (except TX2560 M1, which has DynamicLoM Emulex XE100) there are 2 1Gb Ethernet ports and it is possible to install in a rack.
A very important factor for floor-mounted servers that are located, as a rule, in the work area of office personnel is their noiselessness. Fujitsu PRIMERGY TX servers are very quiet and designed with great attention to the noise factor. If we talk about numbers, in the TX13 series, the server noise is 23 dB (A) in standby mode and 24 dB (A) in operating mode, they are perfect for accounting or operators. The reliability of hardware for PRIMERGY servers is 99.997% (i.e., server downtime is on average 16 minutes per year).
The TX25 series servers are more efficient dual-processor machines. The low-end model, the TX2540 M1, can have up to 192 GB of DDR4 memory. The older model, the TX2560 M1, uses Intel Xeon E5 v3 processors and can have up to 1536 GB of DDR4 memory. These two servers can be used for initial databases or tasks related to virtualization. With the growth of the organization and the emergence of separate server rooms, the equipment can be installed in a rack using a rail-sled.
Fujitsu PRIMERGY BX Blade Servers
In the form factor of the blade servers, Fujitsu today offers two chassis: Fujitsu PRIMERGY BX400 S1 and Fujitsu PRIMERGY BX900 S2. The BX400 S1 is a compact 6U blade server with 8 compartments that can be filled with compute modules, data storage modules, network components. The BX400 S1 chassis can be rack-mounted as well as floor-mounted, on wheels.
This solution has something in common with the quite popular similar solution of Dell VRTX. Fujitsu's main offer in blade server form factor is PRIMERGY BX900 S2. This is a 10U high density chassis with 18 compartments (18 half or 9 full-size). The achieved density indicator of the compartments is currently the first solution available on the market, conditionally conceding only the IBM-Lenovo Flex System x222 solution in density directly to the nodes, which, having 14 compartments, allows you to install an original layout module with 2 independent modules in one compartment servers, which ultimately gives nominally 28 nodes.
Fujitsu PRIMERGY BX400 S1
Currently, three types of computing servers (blades) are available for order: one from the previous generation - BX920 S4, and two from the new generation - BX2560 M1 and BX2580 M1. All blades have two disk compartments and the ability to install up to two processors.
Latest generation blades run on Intel Xeon E5 v3 processors and with DDR4 memory. In the case of a BX2580 M1, up to 1536 GB of RAM can be installed on a node; this model also provides the SATA DOM option, which was described above in the section on PRIMERGY rack servers. The BX2580 M1 is a good option for virtualization-based solutions. Traditionally, much attention is paid to the fault tolerance of the entire chassis: duplicated power supplies, fans, I / O cards (up to 8 pieces).
Fujitsu PRIMERGY BX2560 M1
Fujitsu modular servers PRIMERGY CX
Fujitsu PRIMERGY CX series are special solutions for building high-density horizontal solutions in the form of cluster (or cloud) platforms. The main application of the CX series servers is in the area of cloud computing, High Performance Computing (HPC), deployment of virtualization solutions (for example, Virtual Desktop Infrastructure, Virtual Desktop Infrastructure, VDI). The series is represented by two 2U platforms: CX400 M1 and CX420 S1. The CX400 M1 platform accommodates up to 4 high-performance (dual-processor) nodes and 64 DDR4 memory slots. The CX420 S1 is a cost-optimized solution, a ready-made cluster with two server dual-processor nodes and the ability to install up to 12 shared storage disks, shared power and cooling.
Fujitsu PRIMERGY CX400 M1
If we talk about a more efficient model PRIMERGY CX400 M1, then you can install either 4 dual-processor 1U CX2550 M1 blades (and get 8 E5-2600 v3 processors with up to 18 cores, 144 cores per chassis), or 2 dual-processor 2U CX2570 M1 blades with the option Install up to two full-size PCI cards. For example, NVIDIA GPGPU accelerators or NVIDIA Tesla or Intel Xeon Phi coprocessor boards.
The second model of the CX series - PRIMERGY CX420 S1 - ready cluster, CiB solution (Cluster in a Box). Platform with two 2U server nodes CX272 S1 (2 processors and 2 2.5 ”internal drives in each) with 12 compartments on the platform itself for 2.5” disks used as common storage for both nodes. Such a solution will cost the customer about 20% cheaper in cost than accumulating similar capacity and storage resources with other Fujitsu products - PRIMERGY RX rack servers and ETERNUS DX storage systems, while the volume occupied in the rack will be 2 times smaller.
Before moving on to the Fujitsu PRIMEQUEST line of solutions, let's dilute the technical text with a little historical background. For those for whom such information is not interesting, we recommend skipping this section and continue to the next one.
Fault-tolerant server solutions for critical tasks (the so-called Mission Critical) Fujitsu released for a long time. The manufacturer has been counting since 1956, with the release of the then Fuji Tsushinki Manufacturing Corporation of the first Japanese computer FACOM 128A installed at the Institute of Statistical Mathematics of the Ministry of Education. In 1974, a privately owned company Fujitsu and several others receive state money under a grant from the Ministry of Foreign Trade and Industry (it was this ministry that was actively involved in the program of transition of the national industry from energy-intensive to science-intensive, which fell on the international oil crisis of 1973 ) and announces work on the first M-190 mainframe - the flagship of the M-100 series. In 1975, at the Japan Electronic Computer Show exhibition, many companies presented computers developed by government grants (the state provided active financial support to the national private business in the innovation sector), but attention was focused on the Fujitsu FACOM M-190 because of its performance and compactness for computers LSI (Large-scale integration - technology to integrate thousands of transistors on a single silicon semiconductor microchip used in the period when microprocessors were still in development).
Fujitsu FACOM M-190
A bit of a detective story with oriental notes. Dr. Gene Amdal, a computer architecture guru and author of the law of the same name (Amdahl-Ware law, used to determine the maximum possible increase in the performance of the entire system while increasing the productivity of only part of the system), worked in IBM in the 60s of the last century and was the main architect of the mainframe System / 360 (whose evolution is on the market today under the name IBM zEnterprise System). In 1965, Dr. Amdal was honored with the IBM Fellow title, awarded annually by the corporation’s CEO to four to nine employees, and is the highest recognition of a scientist, engineer, or programmer at IBM. And already in 1970, Dr. Amdal, as the version says, due to IBM’s refusal to accept his ideas on the development of computers, leaves the company and, together with Fujitsu (with participation in capital, ie “for money”), founded Amdahl Corporation. Thus, before receiving a state grant in 1974 and announcing work on the M-190, Fujitsu had been cooperating (and funded) for four years with the former chief architect of IBM mainframes. In the same year of 1975, when the FACOM M-190 made a splash at an exhibition in Japan, the Gina Amdala title corporation presented the Amdahl 470V / 6 mainframe, very much resembling its counterpart from the Japanese co-founder Amdahl Corporation.
The marketing position of Amdahl 470V / 6 was serious: the mainframe was cheaper and more productive than its competitor System / 360 from IBM, and even from Dr. Gene Amdal himself. The first two mainframes went to NASA and the University of Michigan as much as they did, producing an information explosion on both sides of the Pacific. To better understand the sensational event, we recommend recalling the epigraph to the article. Dr. Amdal himself left the company of the same name in 1979 to open a new startup, managing to attract substantial investments in his own name. Since the sale of the first Amdahl 470V / 6 and for the next quarter of a century, Amdahl Corporation (and Fujitsu’s underlying capital) was IBM’s sole competitor in the US mainframe market, reaching a share of 24%, facilitated by the proceedings between IBM and the Justice Department. Antitrust law (analogous to antitrust law in other countries). You can only pay tribute to the Fujitsu strategic managers responsible for entering the US market.
In the mid-1990s, Fujitsu began developing solutions on its own SPARC64 processors (the processor was developed by HAL Computer Systems). A line of PRIMEPOWER servers on SPARC64 processors appears. In 2002, the Japanese manufacturer introduced the mainframe on the SPARC64 - GS21 600 series processors. Fujitsu continues to produce solutions on the SPARC processors, they are available for order (GS21 Series, M-10), but are not so widespread outside the island, although they are quite popular in Japan . In 2005, Fujitsu launches the Fujitsu PRIMEQUEST fault-tolerant servers on an IA-64 microprocessor architecture for mission critical missions (Mission Critical). In cooperation with Fujitsu and using the manufacturer’s PRIMEQUEST servers on the IA-64 architecture, PRIMERGY and Fujitsu ETERNUS storage systems, the arrowhead trading system was developed, implemented and launched in early 2010 on the Tokyo Stock Exchange (TSE). In the same year, the first PRIMEQUEST 1000 series was announced on x86 processors, the current generation of which will be discussed later.
Fujitsu PRIMEQUEST Servers
As mentioned above, Fujitsu PRIMEQUEST servers are solutions designed for mission-critical tasks (Mission Critical). In developing this line, the Japanese manufacturer used its half-century experience and culture of creating mainframes and "hard" solutions, in which special attention is paid to reliability, fault tolerance, non-stop work. If for the Fujitsu PRIMERGY line, the reliability rate is 99.997% (which in itself is very decent for a standard x86 server), then for Fujitsu PRIMEQUEST, the manufacturer declares this figure as the cherished "five nines" - 99.999% (or 5.25 minutes of downtime per year) . Simply put, the fundamental difference between the PRIMERGY servers on the x86 architecture and the PRIMEQUEST servers on the x86 architecture is the increased reliability and resiliency of the latter, the use of the maximum duplication approach and the use of highly reliable components. The consequence of the increased reliability of PRIMEQUEST servers is that PRIMEQUEST servers are quite powerful (the minimum number of supported processors is 4) and large-sized, with the possibility of installing up to 8 processors, a large number of disks, memory bars and expansion cards. This is logical, since there is no point in spending money on achieving the “five nines” of server reliability and applying mass duplication of elements on low-power machines.
The current generation of PRIMEQUEST servers is implemented in the blade server concept; all solutions are implemented in a 10U chassis with the ability to install up to 4 servers. In this case, all installed servers can act as a single server, for example, with 8 processors. As befits mission-critical servers, PRIMEQUEST servers can, without using virtualization platforms using the hardware of the device itself, to divide resources into autonomous partitions (PAR). In this case, each of the 4 physical partitions (PPARs inserted into the server chassis) can be logically divided into several sections of up to 8 pieces (depending on the physical resources of the servers). For specialists working with servers (on other processor architectures) for critical tasks, such functionality is probably familiar. This functionality enables the user of PRIMEQUEST servers, without incurring the financial cost of acquiring a virtualization platform and related software, as well as the loss of server resources for servicing virtualization, using standard tools (hardware and microcode) to allocate resources using extended partitions (eXtended Partition) up to a PCI slot, a processor core and a memory module, or only at the hardware level using physical partitions (PPAR) to the motherboard and I / O devices.
Starting acquaintance directly with the current available solutions in the Fujitsu PRIMEQUEST line, we begin, as is the case with PRIMERGY, with model numbers. Unfortunately, unlike PRIMERGY, the current form of the articles of PRIMEQUEST is not so informative (it is possible that the manufacturer will change it in the future, by analogy with PRIMERGY). The first digit of the article, “2”, means a series of two thousand, this is the current, current, series of PRIMEQUEST servers. Next, the manufacturer divides the servers into two categories: one is called “Business Critical” (simpler, more budget), the second is “Mission Critical” (more powerful, expensive, and with the highest level of reliability). The servers of the Business Critical group in the article have the letter “B” (from “Business”), the servers of the group “Mission Critical” in the article have the letter in the letter not “M”, as suggested, but “E” (from “Enterprise”). The second number in the article, after “2”, indicates the seniority of the model in the series, and the presence of the number after the letter indicates the generation of the model if it is different from the first (in this case, the part number ends in a letter). For example, the name Fujitsu PRIMEQUEST 2400E says that we have a 2000-series server from the Mission Critical group with the maximum implementation of fault tolerance functions (the letter “E”), the youngest in its subgroup (number 4 after 2), model in its first generation. Fujitsu PRIMEQUEST 2800B2 is a two-thousand series, “Business critical”, the oldest model in the group, in the second generation.
Fujitsu PRIMEQUEST 2800B2
As mentioned above, the differences between the “Business Critical” and “Mission critical” groups consist in the mechanisms (their number and functionality) of the RAS (reliability, availability and serviceability) for ensuring the availability of servers. The standard x86 server architecture from the RAS options usually has redundancy for cooling and power systems, RAID protection, error correction code (error-correcting code, ECC) functions and data correction on a single device (Single Device Data Correction, SDDC, algorithm, using ECC to recognize and disable a faulty DRAM device and remove it from the memory card and restore data to a good DRAM). The PRIMEQUST servers of the Business Critical group have all the listed functionality and, moreover, advanced RAS functions, such as: distribution of cache lines to different memory channels (has the names double-device data correction, DDDC, Chipkill, Advanced ECC, Extended ECC , Lockstep Memory), memory mirroring, processor diagnosing hardware errors (Machine Check Architecture, MCA), isolation of the PCIe port upon error detection (PCIe Live Error Recovery, LRE). The servers of the Mission critical (or Enterprise) group have all the above-mentioned functionalities of the previous groups, adding to this the following functions: dynamic reconfiguration (i.e. the ability to add memory, processor, I / O board on the fly), redundancy of the motherboard ), Flexible I/O, .
, PRIMEQUEST . – . (PPAR) , (eXtended Partition), , , -. , , PRIMEQUEST . , PRIMEQUEST PRIMERGY , , PRIMEQUEST RAS .
Fujitsu «», -, : , , , ( ), .
« » . .
- What are you talking about, Doc? All the best gizmos are made in Japan.
- Incredible ...
x / f "Back to the Future"
No less emotionally than the presidency of film actor Ronald Reagan in the “far” 1985, Dr. Emmett Latrop von Braun from 1955 was struck with a remark from the 1985 “boy from the future” Marty McFly that the best electronics in the “far future” produced in Japan. The fact is that before the “economic miracle” of the 70s of the last century, the goods produced in Japan had a low-quality reputation because of the industry destroyed during the Second World War. By the way, a similar stamp of “low-grade junk” in the post-war period was worn in goods made in Germany for exactly the same reasons. And the “engineer Kaiser” was still to put an end to this negligence and to go all the way back to the chased phrase “German quality,” which the Germans did with success.
Presented in the film “Back to the Future-2” by R. Zemeckis, the world of 2015, of course, is different from what we see now behind the window. There were more cars, but they did not take off, but got stuck in traffic jams. Skateboards are still on wheels, although Lexus is trying to “hang” the board with liquid nitrogen. But, probably, if from our 2015, go back to the slightly tuned DeLorean DMC-12, like Marty McFly, 30 years ago, in his native 1985, and show him a broken iPhone 6, Marty will speak out similarly to Doc: “Not surprising that he covered himself. It says, “Made in China.” ”
')
The dominant position of globalization and the struggle of producers for reducing production and logistics costs in the past 30 years have taught us that far from the cheapest products under respectable brands are produced and assembled in countries whose attitude to production capabilities among the masses once was, let's say skeptical. For example, telecommunications and server equipment, as a rule, are produced and assembled in Taiwan, China, India, Mexico, Ireland, Poland, Bulgaria, and the Czech Republic. But despite the domination of globalization and the dictates of cost reduction, there are still manufacturers in our world, whose products contain special and capacious words for the older generation. In this article, we’ll tell you about Fujitsu servers that carry the proud “old school” stamp: “Made in Japan” or “Made in Germany”.
Separately, we note that in view of the geopolitical factor, American manufacturers of IT equipment have stopped selling products to a number of government organizations and companies with state participation, and more recently, to some private companies. The general public is aware that the Chinese manufacturers have been very enthusiastic about these circumstances and wish them good health as long as possible. But it will not be superfluous to know that being under Japanese jurisdiction and, what is more important in this case, based on the strong national support and domestic demand of Fujitsu Limited of the Japanese market, Russian so-called “sanctions” customers are quite loyal.
Fujitsu and x86 architecture
The Japanese company Fujitsu has been manufacturing standard architecture (x86) servers for 20 years. Probably, it can be said that the fact that Fujitsu’s undisputed championship in the national market for server solutions is not the most convincing due to current ideas about the specifics and mentality of the Japanese with a focus on products of national manufacturers, but still it is first in the third largest (in terms of GDP) the economy of the world after the USA and China. In the world market, the company ranks 4th. In other words, Fujitsu is no stranger to the server solutions market and has extensive experience.
The portfolio of Fujitsu standard server solutions (x86) consists of two lines: PRIMERGY and PRIMEQUEST. The PRIMERGY line includes standard rack and tower (floor) servers, as well as Blade servers and CX servers for high performance computing (HPC) and horizontal scaling. The PRIMEQUEST family of servers is focused on high-performance devices based on Intel Xeon E7 with increased component reliability and hardware fault tolerance.
PRIMERGY servers are manufactured at the Fujitsu plant in Augsburg, Germany, while PRIMEQUEST servers are built in Japan, although the possibility of expanding the plant in Bavarian Augsburg and launching an additional PRIMEQUEST assembly line is being considered. Touching on the topic of production, it is worth noting that Fujitsu declares the availability of components to its products for an average of five years from the moment of the official withdrawal of the model from the sale; this is the standard manufacturer's policy. At the same time, the customer has the opportunity to extend the availability of spare parts using a special contract, which will be reserved for the customer. Such an option can be an alternative for those who, due to certain reasons (type of tasks solved by the server, server location, economic reasons), intend to use the server for a sufficiently long period, but do not have the need or the ability to pre-purchase from the warehouse (and for years to withdraw the necessary for storage warehouse space) a list of server spare parts.
Objectively comparing servers of different manufacturers in terms of price / performance ratio, especially in Russia, is quite a difficult task, in the Russian Federation this is complicated by the pricing system used by manufacturers and the sales model, even the exchange selling price for goods may differ depending on the project conditions. Anyone who has had to deal with a marketing presentation of equipment manufacturers more than once knows that all manufacturers “have up to date” any report or review from a specialized organization that shows leadership in various parameters of the manufacturer’s products against direct competitors. As mentioned above, Fujitsu is not the first day on the market and, accordingly, has received such positive reports for itself - for example, from the Transaction Processing Performance Council (TPC), registered in 1988, which specializes in evaluating the performance of computing systems in processing transactions and work with databases. The TPC-E test report is available at the link ; we leave you to assess the credibility of the material.
Fujitsu PRIMERGY Servers
In the line, as mentioned above, the servers are rack and tower versions. The model number should be read as follows: the letter R, T or B indicates the server form factor, respectively, rack (R - rack), tower (T - tower) or blade server (B - blade). Next comes a slight (temporary) confusion. Articles of the previous generation of Fujitsu PRIMERGY servers consisted of a three-digit numeric number and an “S” at the end. The current generation has a four-digit numerical number in the article and “M1” at the end. The three-digit article of the servers of the previous generation today can only tell the user that this is the previous generation and the server is obsolete, whereas in the new generation the manufacturer has made the articles more informative. The first digit of the number indicates the possible number of processors (socket) in the server. The second digit is the series number of the used Intel Xeon processors, i.e. E3, E5, E7 (respectively, "3", "5", "7"). For example: RX4770 M1 is a rack server with four Intel Xeon E7 sockets, TX2540 M1 is a tower server with two Intel Xeon E5 sockets. In our opinion, the manufacturer offered a fairly convenient model for marking devices, which allows without going into the specification to “read” some of the basic characteristics of the server according to the article, and sometimes avoid extremely annoying (and expensive) errors.
A number of PRIMERGY server models (mostly older models) support, in their series, technologies for the developer and manufacturer of network solutions from Emulex, including Emulex Universal Multi-Channel technology (UMC), which allows you to create multiple PCI functions on each network adapter. For example, each 10GbE port can be configured with 8 functions of the virtual network adapter. The user has the ability to configure the bandwidth of each channel and traffic separation, the entire system will work correctly with any 10GbE switch.
As already noted, the Japanese manufacturer in the design of PRIMERGY servers paid great attention to the architecture and layout of the servers. The servers are not only energy efficient (unfortunately, we often don’t care about this customer’s indicator) and are quiet (this indicator is much more interesting), but they can also work 24/7 at temperatures up to 40 ° C.
Fujitsu PRIMERGY RX Rack Servers
At present, the PRIMERGY server rack group of the new generation includes: single-unit servers RX1330 M1 (with one Intel Xeon E3 processor) and RX2530 M1 (up to two processors). Two dual-module dual-processor models RX2520 M1 and RX2540 M1, two four-unit models RX2560 M1 and RX4770 M1 (with the possibility of installing up to 4 processors). Also currently available to order servers from the previous generation: dual-dual dual-processor RX300 S8 and four-dual dual-processor RX350 S8.
Fujitsu PRIMERGY RX2520 M1
The differences between the servers are standard: the number of installed processors and the power consumed by them, the number of slots and the maximum amount of RAM, disks, the number of installed PCIe cards, etc.
In addition to the two low-end models, the RX1330 M1 and RX2520 M1, all Fujitsu PRIMERGY rack-mount servers, including the single-unit RX2530 M1, support expandable RAM up to 1536 GB or more (in 4U models). The servers use only disks of large (LFF, 3.5 ″) and small (SFF, 2.5 ″) form factors, the manufacturer uses 1.8 ″ disks only in blade servers. Some models of Fujitsu PRIMERGY rack servers have the ability to install PCIe-SSD disks, which are becoming more common for solving problems that require fast SSD disks for caching (for example, in the implementation of the so-called Software Defined solutions).
In its servers (high models), Fujitsu offers an option that is not often seen from other manufacturers - the ability to install internal DOM (Disk on Module) drives with SATA 3G interface into the SATA connector directly on the motherboard. Rather inexpensive and energy-efficient drives, with reliability significantly higher than that of a traditional USB flash drive or SD card. Such drives are designed primarily to perform the functions of a boot disk and are an alternative solution in a common scheme with SD cards (single or duplicated).
Also, older models support DynamicLoM technology. This solution, which essentially pre-installs a network adapter at a speed of up to 40 Gbps onboard the server's motherboard, while the user chooses the way out based on his needs. That is, you can use 1 Gbps output and modify it to 10 or 40 Gbps (in the near future), depending on the needs.
Fujitsu PRIMERGY TX Tower Servers
The PRIMERGY tower server group consists of three single-processor servers TX1310 M1, TX1320 M1, TX1330 M1 and two dual-processor servers TX2540 M1, TX2560 M1. From the previous generation, the single-processor TX150 S8 and the dual-processor TX300 S8 are still available for order.
Fujitsu PRIMERGY TX2560 M1
Fujitsu TX tower servers, depending on the model, can work with 1 or 2 Intel Xeon E3 and E5 processors. Uses large (LFF, 3.5 ") and small (SFF, 2.5") form factors. In all servers (except TX2560 M1, which has DynamicLoM Emulex XE100) there are 2 1Gb Ethernet ports and it is possible to install in a rack.
A very important factor for floor-mounted servers that are located, as a rule, in the work area of office personnel is their noiselessness. Fujitsu PRIMERGY TX servers are very quiet and designed with great attention to the noise factor. If we talk about numbers, in the TX13 series, the server noise is 23 dB (A) in standby mode and 24 dB (A) in operating mode, they are perfect for accounting or operators. The reliability of hardware for PRIMERGY servers is 99.997% (i.e., server downtime is on average 16 minutes per year).
The TX25 series servers are more efficient dual-processor machines. The low-end model, the TX2540 M1, can have up to 192 GB of DDR4 memory. The older model, the TX2560 M1, uses Intel Xeon E5 v3 processors and can have up to 1536 GB of DDR4 memory. These two servers can be used for initial databases or tasks related to virtualization. With the growth of the organization and the emergence of separate server rooms, the equipment can be installed in a rack using a rail-sled.
Fujitsu PRIMERGY BX Blade Servers
In the form factor of the blade servers, Fujitsu today offers two chassis: Fujitsu PRIMERGY BX400 S1 and Fujitsu PRIMERGY BX900 S2. The BX400 S1 is a compact 6U blade server with 8 compartments that can be filled with compute modules, data storage modules, network components. The BX400 S1 chassis can be rack-mounted as well as floor-mounted, on wheels.
This solution has something in common with the quite popular similar solution of Dell VRTX. Fujitsu's main offer in blade server form factor is PRIMERGY BX900 S2. This is a 10U high density chassis with 18 compartments (18 half or 9 full-size). The achieved density indicator of the compartments is currently the first solution available on the market, conditionally conceding only the IBM-Lenovo Flex System x222 solution in density directly to the nodes, which, having 14 compartments, allows you to install an original layout module with 2 independent modules in one compartment servers, which ultimately gives nominally 28 nodes.
Fujitsu PRIMERGY BX400 S1
Currently, three types of computing servers (blades) are available for order: one from the previous generation - BX920 S4, and two from the new generation - BX2560 M1 and BX2580 M1. All blades have two disk compartments and the ability to install up to two processors.
Latest generation blades run on Intel Xeon E5 v3 processors and with DDR4 memory. In the case of a BX2580 M1, up to 1536 GB of RAM can be installed on a node; this model also provides the SATA DOM option, which was described above in the section on PRIMERGY rack servers. The BX2580 M1 is a good option for virtualization-based solutions. Traditionally, much attention is paid to the fault tolerance of the entire chassis: duplicated power supplies, fans, I / O cards (up to 8 pieces).
Fujitsu PRIMERGY BX2560 M1
Fujitsu modular servers PRIMERGY CX
Fujitsu PRIMERGY CX series are special solutions for building high-density horizontal solutions in the form of cluster (or cloud) platforms. The main application of the CX series servers is in the area of cloud computing, High Performance Computing (HPC), deployment of virtualization solutions (for example, Virtual Desktop Infrastructure, Virtual Desktop Infrastructure, VDI). The series is represented by two 2U platforms: CX400 M1 and CX420 S1. The CX400 M1 platform accommodates up to 4 high-performance (dual-processor) nodes and 64 DDR4 memory slots. The CX420 S1 is a cost-optimized solution, a ready-made cluster with two server dual-processor nodes and the ability to install up to 12 shared storage disks, shared power and cooling.
Fujitsu PRIMERGY CX400 M1
If we talk about a more efficient model PRIMERGY CX400 M1, then you can install either 4 dual-processor 1U CX2550 M1 blades (and get 8 E5-2600 v3 processors with up to 18 cores, 144 cores per chassis), or 2 dual-processor 2U CX2570 M1 blades with the option Install up to two full-size PCI cards. For example, NVIDIA GPGPU accelerators or NVIDIA Tesla or Intel Xeon Phi coprocessor boards.
The second model of the CX series - PRIMERGY CX420 S1 - ready cluster, CiB solution (Cluster in a Box). Platform with two 2U server nodes CX272 S1 (2 processors and 2 2.5 ”internal drives in each) with 12 compartments on the platform itself for 2.5” disks used as common storage for both nodes. Such a solution will cost the customer about 20% cheaper in cost than accumulating similar capacity and storage resources with other Fujitsu products - PRIMERGY RX rack servers and ETERNUS DX storage systems, while the volume occupied in the rack will be 2 times smaller.
Fujitsu mainframe development history
Before moving on to the Fujitsu PRIMEQUEST line of solutions, let's dilute the technical text with a little historical background. For those for whom such information is not interesting, we recommend skipping this section and continue to the next one.
Fault-tolerant server solutions for critical tasks (the so-called Mission Critical) Fujitsu released for a long time. The manufacturer has been counting since 1956, with the release of the then Fuji Tsushinki Manufacturing Corporation of the first Japanese computer FACOM 128A installed at the Institute of Statistical Mathematics of the Ministry of Education. In 1974, a privately owned company Fujitsu and several others receive state money under a grant from the Ministry of Foreign Trade and Industry (it was this ministry that was actively involved in the program of transition of the national industry from energy-intensive to science-intensive, which fell on the international oil crisis of 1973 ) and announces work on the first M-190 mainframe - the flagship of the M-100 series. In 1975, at the Japan Electronic Computer Show exhibition, many companies presented computers developed by government grants (the state provided active financial support to the national private business in the innovation sector), but attention was focused on the Fujitsu FACOM M-190 because of its performance and compactness for computers LSI (Large-scale integration - technology to integrate thousands of transistors on a single silicon semiconductor microchip used in the period when microprocessors were still in development).
Fujitsu FACOM M-190
A bit of a detective story with oriental notes. Dr. Gene Amdal, a computer architecture guru and author of the law of the same name (Amdahl-Ware law, used to determine the maximum possible increase in the performance of the entire system while increasing the productivity of only part of the system), worked in IBM in the 60s of the last century and was the main architect of the mainframe System / 360 (whose evolution is on the market today under the name IBM zEnterprise System). In 1965, Dr. Amdal was honored with the IBM Fellow title, awarded annually by the corporation’s CEO to four to nine employees, and is the highest recognition of a scientist, engineer, or programmer at IBM. And already in 1970, Dr. Amdal, as the version says, due to IBM’s refusal to accept his ideas on the development of computers, leaves the company and, together with Fujitsu (with participation in capital, ie “for money”), founded Amdahl Corporation. Thus, before receiving a state grant in 1974 and announcing work on the M-190, Fujitsu had been cooperating (and funded) for four years with the former chief architect of IBM mainframes. In the same year of 1975, when the FACOM M-190 made a splash at an exhibition in Japan, the Gina Amdala title corporation presented the Amdahl 470V / 6 mainframe, very much resembling its counterpart from the Japanese co-founder Amdahl Corporation.
The marketing position of Amdahl 470V / 6 was serious: the mainframe was cheaper and more productive than its competitor System / 360 from IBM, and even from Dr. Gene Amdal himself. The first two mainframes went to NASA and the University of Michigan as much as they did, producing an information explosion on both sides of the Pacific. To better understand the sensational event, we recommend recalling the epigraph to the article. Dr. Amdal himself left the company of the same name in 1979 to open a new startup, managing to attract substantial investments in his own name. Since the sale of the first Amdahl 470V / 6 and for the next quarter of a century, Amdahl Corporation (and Fujitsu’s underlying capital) was IBM’s sole competitor in the US mainframe market, reaching a share of 24%, facilitated by the proceedings between IBM and the Justice Department. Antitrust law (analogous to antitrust law in other countries). You can only pay tribute to the Fujitsu strategic managers responsible for entering the US market.
In the mid-1990s, Fujitsu began developing solutions on its own SPARC64 processors (the processor was developed by HAL Computer Systems). A line of PRIMEPOWER servers on SPARC64 processors appears. In 2002, the Japanese manufacturer introduced the mainframe on the SPARC64 - GS21 600 series processors. Fujitsu continues to produce solutions on the SPARC processors, they are available for order (GS21 Series, M-10), but are not so widespread outside the island, although they are quite popular in Japan . In 2005, Fujitsu launches the Fujitsu PRIMEQUEST fault-tolerant servers on an IA-64 microprocessor architecture for mission critical missions (Mission Critical). In cooperation with Fujitsu and using the manufacturer’s PRIMEQUEST servers on the IA-64 architecture, PRIMERGY and Fujitsu ETERNUS storage systems, the arrowhead trading system was developed, implemented and launched in early 2010 on the Tokyo Stock Exchange (TSE). In the same year, the first PRIMEQUEST 1000 series was announced on x86 processors, the current generation of which will be discussed later.
Fujitsu PRIMEQUEST Servers
As mentioned above, Fujitsu PRIMEQUEST servers are solutions designed for mission-critical tasks (Mission Critical). In developing this line, the Japanese manufacturer used its half-century experience and culture of creating mainframes and "hard" solutions, in which special attention is paid to reliability, fault tolerance, non-stop work. If for the Fujitsu PRIMERGY line, the reliability rate is 99.997% (which in itself is very decent for a standard x86 server), then for Fujitsu PRIMEQUEST, the manufacturer declares this figure as the cherished "five nines" - 99.999% (or 5.25 minutes of downtime per year) . Simply put, the fundamental difference between the PRIMERGY servers on the x86 architecture and the PRIMEQUEST servers on the x86 architecture is the increased reliability and resiliency of the latter, the use of the maximum duplication approach and the use of highly reliable components. The consequence of the increased reliability of PRIMEQUEST servers is that PRIMEQUEST servers are quite powerful (the minimum number of supported processors is 4) and large-sized, with the possibility of installing up to 8 processors, a large number of disks, memory bars and expansion cards. This is logical, since there is no point in spending money on achieving the “five nines” of server reliability and applying mass duplication of elements on low-power machines.
The current generation of PRIMEQUEST servers is implemented in the blade server concept; all solutions are implemented in a 10U chassis with the ability to install up to 4 servers. In this case, all installed servers can act as a single server, for example, with 8 processors. As befits mission-critical servers, PRIMEQUEST servers can, without using virtualization platforms using the hardware of the device itself, to divide resources into autonomous partitions (PAR). In this case, each of the 4 physical partitions (PPARs inserted into the server chassis) can be logically divided into several sections of up to 8 pieces (depending on the physical resources of the servers). For specialists working with servers (on other processor architectures) for critical tasks, such functionality is probably familiar. This functionality enables the user of PRIMEQUEST servers, without incurring the financial cost of acquiring a virtualization platform and related software, as well as the loss of server resources for servicing virtualization, using standard tools (hardware and microcode) to allocate resources using extended partitions (eXtended Partition) up to a PCI slot, a processor core and a memory module, or only at the hardware level using physical partitions (PPAR) to the motherboard and I / O devices.
Starting acquaintance directly with the current available solutions in the Fujitsu PRIMEQUEST line, we begin, as is the case with PRIMERGY, with model numbers. Unfortunately, unlike PRIMERGY, the current form of the articles of PRIMEQUEST is not so informative (it is possible that the manufacturer will change it in the future, by analogy with PRIMERGY). The first digit of the article, “2”, means a series of two thousand, this is the current, current, series of PRIMEQUEST servers. Next, the manufacturer divides the servers into two categories: one is called “Business Critical” (simpler, more budget), the second is “Mission Critical” (more powerful, expensive, and with the highest level of reliability). The servers of the Business Critical group in the article have the letter “B” (from “Business”), the servers of the group “Mission Critical” in the article have the letter in the letter not “M”, as suggested, but “E” (from “Enterprise”). The second number in the article, after “2”, indicates the seniority of the model in the series, and the presence of the number after the letter indicates the generation of the model if it is different from the first (in this case, the part number ends in a letter). For example, the name Fujitsu PRIMEQUEST 2400E says that we have a 2000-series server from the Mission Critical group with the maximum implementation of fault tolerance functions (the letter “E”), the youngest in its subgroup (number 4 after 2), model in its first generation. Fujitsu PRIMEQUEST 2800B2 is a two-thousand series, “Business critical”, the oldest model in the group, in the second generation.
Fujitsu PRIMEQUEST 2800B2
As mentioned above, the differences between the “Business Critical” and “Mission critical” groups consist in the mechanisms (their number and functionality) of the RAS (reliability, availability and serviceability) for ensuring the availability of servers. The standard x86 server architecture from the RAS options usually has redundancy for cooling and power systems, RAID protection, error correction code (error-correcting code, ECC) functions and data correction on a single device (Single Device Data Correction, SDDC, algorithm, using ECC to recognize and disable a faulty DRAM device and remove it from the memory card and restore data to a good DRAM). The PRIMEQUST servers of the Business Critical group have all the listed functionality and, moreover, advanced RAS functions, such as: distribution of cache lines to different memory channels (has the names double-device data correction, DDDC, Chipkill, Advanced ECC, Extended ECC , Lockstep Memory), memory mirroring, processor diagnosing hardware errors (Machine Check Architecture, MCA), isolation of the PCIe port upon error detection (PCIe Live Error Recovery, LRE). The servers of the Mission critical (or Enterprise) group have all the above-mentioned functionalities of the previous groups, adding to this the following functions: dynamic reconfiguration (i.e. the ability to add memory, processor, I / O board on the fly), redundancy of the motherboard ), Flexible I/O, .
, PRIMEQUEST . – . (PPAR) , (eXtended Partition), , , -. , , PRIMEQUEST . , PRIMEQUEST PRIMERGY , , PRIMEQUEST RAS .
Fujitsu «», -, : , , , ( ), .
« » . .
Source: https://habr.com/ru/post/267633/
All Articles