TW200819950A - Blade clustering system with SMP capability and redundant clock ditribution architecture thereof - Google Patents

Abstract

A redundant clock distribution architecture is provided for a blade clustering system to achieve SMP (symmetric multi-processor) capability and flexible system configuration. The architecture mainly provides a central clock signal from a central clock and a local clock signal from an operative local clock configured on each blade module of the system. A clock multiplexer selects the central clock signal and sends to plural local clock consumers on each blade module. The clock multiplexer switches to send the local clock signal if the central clock fails.

Description

200819950 IX. INSTRUCTIONS: [Technical Field] The present invention relates to a timing allocation architecture in a computer system, particularly an alternate timing allocation architecture for a blade-type cluster system having symmetric multi-processor performance. [Prior Art] At present, the substantial difference between the 'SMP system' and the blade clustering system is still difficult to break. These two systems and charges are for a certain. The purpose is to design with operation in a certain configuration. To combine two types of systems into one, timing allocation will be one of the main problems. In general, a 'reliable and high-order symmetric multiprocessor system has an alternate timing distribution system' and a typical blade-type system has a turn-domain timing generation and distribution system on each blade. The first figure is an example of a wire-symmetric multi-site system for prior art towel tree ordering. The system with a plurality of computing modules 12〇, 13〇, 14〇... includes a first central timing device Chuan φ and a younger day shouting 2112 as the same time source of the same, which exchanges timing edge correction information (clock edge Ahgnment infGrmatiQn) to maintain the same timing edge correction. Two or _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ Each of 130, 140 receives two replicated signals that are sub-allocated by the first central timingr m and the second central timing. The multiplexer (2)/(3)/(4) is used for the magnetic-office, if one of the -_central timing or the second central timing fails, the other is used. Therefore, a typical symmetrical multiprocessor system usually needs to be stepped 5 200819950 (4) or day _ edge correction can be handled. In addition, the region sequencer 122/132/142 may be used only for testing. Because the region sequencer (2) my(4) does not have the same/or the day t sequence, it also grabs the 'under normal operation' region timing device (2) / 132 / 142 is generally disabled ulsa_) or suspended (suspended). Therefore, the regional sequencer is a waste of the performance of the symmetric multiprocessing system in the space of the substrate. The second picture shows the example of the technique of the blade-type system (10). Because each blade is off, 153..·妓—operational rhyme, _ vertical (4), this does not need to be thin. Over-synchronization or timing edge correction, edge - The area sequencer of the knife box, 152, 153.·· is out of order, and other blades _ continue to operate. As long as -_f turn 仏丨 _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ Therefore, the first _ most of the knives "do not have _ sequencer (she phase c_)). On the other hand, it is also believed that the blade system does not have the performance of multiple symmetry in the blade because it lacks synchronization or timing edge correction. In order to achieve a flexible system configuration, the blade system has the performance of a symmetric multi-processor. It is first necessary to rearrange the timing generation and allocation architecture. SUMMARY OF THE INVENTION The present invention provides an alternate timing allocation architecture that enables a blade cluster system to have symmetric multiple energy. The _-central_and fine_domain timing devices, while the knives clustering system has large symmetrical multiprocessor performance. In the embodiment of the present invention, the symmetric multiprocessor performance ^, the annual cluster system includes a side blade module and a timing distribution architecture. Each blade module contains multiple regional timings for 200819950 - household. The timing allocation architecture includes a timing sequence 111 (ClQCk multipIexer), an operational area sequencer disposed on each of the ride modules, and a command timing sequencer. The central timing device is used to generate the phase number of the towel, the timing signal of the messenger and the regional timing signal generated by the _timer, and the central timing signal is selected by the timing multiplexer and sent to each blade modulo _ When the central phase is out of order, the time-series multi-worker on the blade module switches to the t-time sequence and sends the sequence signal to the regional time series user. [Embodiment] > The present invention provides a timing architecture that enables a blade-type device to have a high-order symmetry township processing 1 and has a _ _ system structure. - Blade-type cluster system mainly includes - central timing (4) and multiple blade modules 20, 30, L-blade clusters basically contain the necessary hardware damage and a cluster management system (not shown) to manage the blade module 2〇, 3〇, 4〇, 5〇 operation. (In order to provide a clearer explanation, the other system components included in the blade cluster system are omitted in the drawing.) The cluster management system is a software program, which is in the group 2〇, 3〇, 4〇, 5〇 Use and manage as a management/operation interface. In the present invention, the cluster management system supports cluster computing (cluster C0_ing), including asy deleting etric clustering with and without a master node, and symmetric clustering (_etric_fine ★). Each blade module 20/30/40/50 can be used as a stand-alone computer, usually on a motherboard. Blade module 2G/3G/4G/50 with different electronic components such as: processor, system memory, body, bridge chip, input / wheel controller, network interface controller, expansion card input / wheel. Connectors (all not shown), a timing multiplexer (MUX) 21/31741/51, and region timing p 22/32/42/52' are embodied on printed circuit boards. These components are connected via a specific busbar. 7 200819950 The 2G/3_/5G is used to perform direct monitoring and management of the hardware components on which it is set up, to manage and provide application software. In the present invention, each job dedicated to the blade module has the performance of performing a symmetric multiprocessor configuration. The blade unit 2〇, go, “n # slot (not shown) of the single processor (not shown) is installed in the dedicated processor · 曰曰 film, the mother is set with - or more计曾桉~(7)嗔lng called Qiqi invention, in riding _ = support various symmetrical multi-processor configurations, such as small 2, 4 (four) that is, 'do-like jing' editorial area. Set (4) to touch the outside or set in the center 10 of the blade module 2G, 3q, Xian, evil - to generate and distribute - synchronous financial timing "for each - knife front 2G, 3G, 4G, only 5 / in some mosquitoes For example, the central timing signal can be provided by the regional timing signal of the blade modules 2〇, 30, 40, 5(). Central timing cluster task. ^1G疋 is controlled by the cluster smoke system to perform the setting of each __ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ Operating area timing signal. The area sequencer 22/3_2 also operates early, such as: testing, debugging, problem repair, time zone phase assist, pure state can maintain _ timing edge correction, blade cluster (10) can be single-blade operation (also coffee Coffee e W mail (8) money complete (four) sequence transfer to Wei Li domain sign (see four pictures). Time series multiplexer 21/31/41/51 electrical continuous shooting phase sequencer 1 and in each - blade module 200819950 20/30/40 /50 area sequencer 22/32/42/52 and area timing users. The central timing signal and area timing signals are sent to the timing multiplexer 21/31/41/51.

The day-to-day state 21/31/41/51 monitors the state of the timing signal, and a sound timing signal can be selected from the central timing signal and the regional day-sequence signal. The timing multiplexer 2]/31/41/51 can be preset to select the central timing signal. If the selected central timing signal has a problem, and other timing signals are sound, the timing is 4 41/31/41/5 〗 JIANG _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ The cluster system monitors the timing of the sequence of the completion of the sequence, and takes the necessary actions to restore the blade cluster system. Please refer to the fourth figure. In the single-blade mode (smgle-blade mode), there is only one blade module 2〇/3〇/4〇/5〇 and each operating system 〇2/_ full Covered by the field. In this mode, Mengsheng and distribution are completely spare. Like the - cluster system, the blade cluster system can be preset as the centimeter timing signal. Once the central sequencer 1G fails or has a problem, the timing multiplexer 21/31/41/51 switches and the blade cluster is charged. The Haikufeng core group 20/30/40/50 can use the regional timing signal generated by its regional sequencer 22/32/42/52, which is the operation of the simple blade module _ /5/5q. If the area sequencer 22/32/42/52 can guarantee the secret of the area, the 1st k-field % sequence and the central time-series signal are at the center of the signal failure _ timing edge correction, the entire blade can be bribed Synchronous clusters operate relatively. If there is no synchronized regional timing signal, each blade module will continue to operate. As long as the cluster management system continues to handle task scheduling/dispatch, the blade cluster system can still be used for new ones. task. 200819950 Please refer to the fifth figure' In the multi-blade mode (muitiple-blad(9)peratiGnm〇de), the processor of the two-blade module (20, 30) M40, 50) is connected by two system bus bars.

To form two or more symmetric multiprocessor / operating system fields 〇 6, 〇 7. A network connection connects the symmetric multiprocessor/operating system domain〇6,〇7 to clustering' This configuration needs to be synchronized in the same-symmetric multiprocessor/operation system domain Q6, Q7 Timing. In an embodiment, if the central timing device 1G fails, the region timing devices (2), 32) / (42, 52) cannot provide processor synchronization timing signals in the same-symmetric multiprocessor/operation system field. Tasks in progress will not reply. In the present day, the system bus can be arbitrarily multi-processed by any connection and multiple processing circuits. For example, such bus bars depend on the Transp〇rt protocol. The network link contains the actual high-speed interface between the network interface controllers connected to the blade module, for example: Infinite Broadband Link (lnfmite Band _11) or Gigabit Ethernet Link (Glgabyte)

Ethernet connection) 〇 One solution is - the synchronization module (not shown) turns off the copper _ to the multi-processor / operating system area _ the area sequencer (2), take (4), in). "Another--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- The order comes back, counts, and; s, eve & bury, and restarts the system in the early blade operation mode. Therefore, the blade does not need any repair or replacement, and I will replace it in the prior art. Or repairing the hardware configuration is very time consuming. Using the timing hardware μ. 1 structure 'The present invention provides the opportunity for the blade-type cluster to maintain a stable and continuous period of operation. 守连作10 200819950 t sequence of 7 pre-. Full standby mode (fuH-redundant m〇de). If for some reason the scale cluster system cannot operate in the single-blade mode, the cluster management system will need to re-power it, change the timing source and then re- Start the system, just like the single blade operation mode. The ten pairs of the external order source are used as the central timing of those blades, in the test those blades will be called the surface of the face. This invention provides Xiangdian test, debugging and problem repair Elasticity, knife The front can therefore operate as a stand-alone computer. The hardware and/or software/web configuration of the mains will probably need to be used to change the symmetrical multiprocessing 4 configuration. One of the basic conditions for the configuration of the elastic line. The present invention has been disclosed in the preferred embodiment above, and the financial is not limited to the spirit - a little change with the spirit

In the present invention, any person skilled in the art, without departing from the invention, should be covered by the invention, and the scope of protection of the present invention is defined by the scope of the patent application. Prevail. [Simple Description of the Drawings] | Embodiments f is a timing distribution of a high-order symmetric multiprocessor system in the prior art. The second figure is a typical device of the prior art bladed cluster. The third figure shows that the blade-type Ye Yingchong with the alternate timing architecture according to the embodiment of the present invention is the blade-shaped W, ,, '. shown in the third figure. The 硪 集 减 于 于 in the single-blade mode. The five-figure operation mode of the blade operation mode is set to the blade-type cluster system shown in the third figure. [Main component symbol description] 〇2, 03, 04, 05·• Operation wire 11 200819950 06, 07 ··Symmetric multi-processing / Operation System Field 10: Central Sequencer 111: First Central Timer 112: Second Central Timer 120, 130, 140: Computing Modules 121, 131, 141: Multiplexers 122, 132, 142: Area Timing • I 150: Blade-type cluster system 151, 152, 153: Blades 20, 30, 40, 50: Blade modules 21, 31, 41, 51: Timing multiplexers 22, 32, 42, 52 · Area timing #12

Claims (1)

200819950 Ten '申凊 patent scope: 1. A Na Na 峨 __ fine front cluster system contains a plurality of blade modules, the timing distribution architecture = structure, the knife a central phase, generate - central timing signal; and timing ^^―Butterfly touches, shirt steals auxiliary ^ 4 receives the central timing signal and the region sequencer produces - μ _ =: ~ 喻 职 伽 伽 = = = = = = = = = = = = = = The timing multi-worker on the blade module switches the central timing signal to send the regional timing signal to the time series users in the area. 2. If the timing of the request item 1 is scaled, the (four) central timing signal is such as a knife-shaped mosquito. One of the region's sequencers is provided. ~ 3. If the time series allocation _ of the request item, further includes the -_ group, use the step-by-step method to process the timing signals of the two or more of the blade modules in at least the symmetry . 4. The timing allocation architecture of claim 3, wherein in the same-symmetric multiprocessor range, the complex processing (four) on the different blade modules is connected by a system and a surface stream. 5. The timing allocation architecture of claim 1, wherein the blade cluster system further comprises a network connection between the blade modules. 6. A blade-shaped green edge with _«News, including: a plurality of blades, a phase-sharing _ time phase; and a timing distribution architecture comprising: a central timing device to generate a central timing signal; 13 200819950 - Timing multi-guard and - operable area sequencer, installed in each of the blade modules, the timing is more than 4, and the area timing signal generated by the timing sequence and the area sequencer is more than that timing. Selecting the central timing job and sending to the regional timing users of each blade module; wherein 'if the oblique phase sequencer fails, the timing multiplexer located on the blade module converts the central timing call to Send the area timing signal to the time series users = The blade-end clustering system of claim 6, wherein the central timing signal is provided by the region sequencer of one of the blade modules. δ. The blade-type programming of claim 6, wherein the timing distribution architecture further comprises a - synchronous mode, and "using at least one symmetric multiprocessor fan, two or more of the blades The timing signal of the area of the module. 9. The blade system of claim 8 is the most wood system, wherein in the same symmetric multiprocessor, the cliffs on the blade modules are different. Processing _ wrong by -1 dragon flow row connection. 10 · If the request of the 6 blade of the industry 隹 # 丛 wood,, charge, and a network link between the blade module.