HK1164579B

HK1164579B - Method for managing a computing center and managing an energy efficient control policy in a computing center resource resident in a computing center

Info

Publication number: HK1164579B
Application number: HK12105013.8A
Authority: HK
Inventors: 韦尔.威廉.戴博; 大卫．贝瑞
Original assignee: 美国博通公司
Priority date: 2010-04-07
Filing date: 2012-05-22
Publication date: 2015-07-10

Abstract

The present invention is directed to a method for managing a computing center in a management station and a method for managing an energy efficient control policy in a computing center resource resident in a computing center. It relates to energy efficient Ethernet (EEE) enhanced information technology power management tools. EEE-based computing center resources are designed to monitor energy savings events hardware components (e.g., physical layer device) included within the computing center resource. Energy saving statistics based on such monitoring can be provided to a power management tool. This monitoring information enables the power management tool to make broad service-level energy savings decisions on actual network activity. In addition, feedback based on the broad service-level energy savings decisions can be provided to the EEE-based computing center resources for consideration by their individual EEE control policies.

Description

Management computing center and method for managing high-energy-efficiency control strategy of computing center resources

Technical Field

The present invention relates to power management, and more particularly to a system and method for an energy efficient ethernet enhanced information technology power management tool.

Background

In recent years, there is a tendency of increasing energy costs. Accordingly, various industries are increasingly sensitive to the impact caused by this increasing cost. An increasingly scrutinizing component is IT facilities. Many companies are currently researching the power usage of their IT systems to determine whether energy costs can be reduced.

Efforts to eliminate redundant power usage are very valuable to data centers and computing centers, which have a large number of server systems in their homes. By cutting off unneeded servers and reusing (reusing) servers on demand is one of the energy savings ways. With these power saving techniques, the fact that ideal matching of computing resources and computing requirements can minimize power consumption has been recognized. Therefore, there is a need for a mechanism that enables IT administrators to be provided with a power management tool that makes them intelligent power saving decisions.

Disclosure of Invention

According to one aspect of the present invention, there is provided a method of managing a computing center in a management station, the computing center comprising a plurality of computing center resources for providing a plurality of computing services, the method comprising:

the management station receiving, from a first compute hub resource of the plurality of compute hub resources, first energy savings information for a first physical layer device of the first compute hub resource, the first energy savings information generated by the first compute hub resource based on a first energy savings state used by the first physical layer device in response to a first low traffic utilization condition detected by the first compute hub resource;

the management station receiving, from a second compute hub resource of the plurality of compute hub resources, second energy savings information for a second physical layer device of the second compute hub resource, the second energy savings information generated by the second compute hub resource based on a second energy savings state used by the second physical layer device in response to a second low traffic utilization detected by the second compute hub resource;

determining a utilization level of a computing service provided by the computing center based on the received first and second energy saving information; and

based on the decision, shutting off (switch off) compute farm resources for providing the compute service.

Preferably, the receiving comprises receiving from a switch.

Preferably, the receiving comprises receiving from a server.

Preferably, the receiving comprises receiving statistics of physical layer devices entering a power saving state.

Preferably, said switching off comprises switching off a server.

Preferably, the cutting comprises reusing (repurpose) the server.

According to one aspect of the present invention, there is provided a method of managing a computing center in a management station, the management center comprising a plurality of computing center resources for providing a plurality of computing services, the method comprising:

the management station receiving, from a first compute hub resource of the plurality of compute hub resources, first energy saving information for a first hardware subsystem of the first compute hub resource, the first energy saving information generated by the first compute hub resource based on a first energy saving state used by the first hardware subsystem;

the management station receiving, from a second compute hub resource of the plurality of compute hub resources, second energy saving information for a second hardware subsystem of the second compute hub resource, the second energy saving information generated by the second compute hub resource based on a second energy saving state used by the second hardware subsystem;

based on the determination, shutting down computing center resources for providing the computing service.

Preferably, the receiving comprises receiving from a switch.

Preferably, the receiving comprises receiving from a server.

Preferably, the receiving comprises receiving statistics of hardware subsystems entering a power saving state.

Preferably, said switching off comprises switching off a server.

Preferably, said switching off comprises reusing a server.

According to one aspect of the present invention, there is provided a method of managing an energy-efficient control strategy for a computing center resource, wherein the computing center resource is at a computing center, the method comprising:

monitoring that a hardware subsystem in a first computing center resource enters an energy-saving state, wherein the hardware subsystem enters the energy-saving state and is controlled by an energy-efficient control strategy;

the compute farm resource transmits energy-saving statistics to a management station based on the monitoring, the management station managing a plurality of compute farm resources, the plurality of compute farm resources including the first compute farm resource;

the management station determines the utilization level of the computing service provided by the computing center by using the energy-saving statistics transmitted by the plurality of computing center resources;

the management station transmitting utilization information to the first computing resource based on the decision; and

changing an energy-efficient control strategy in the first compute farm resource based on the transmitted utilization information.

Preferably, the monitoring comprises monitoring by a server.

Preferably, the monitoring comprises monitoring that physical layer devices in the first compute farm resource enter the power save state.

Preferably, transmitting the power-saving statistics comprises transmitting an amount of time that the hardware subsystem remains in the power-saving state during a measurement interval.

Preferably, the changing comprises changing a level of aggressiveness of the energy-efficient control strategy in the first computing center resource.

Drawings

In order to describe the manner in which the above-recited and other advantages and features of the invention can be obtained, a more particular description of the invention briefly described above will be rendered by reference to specific embodiments thereof which are illustrated in the appended drawings. Understanding that these drawings depict only typical embodiments of the invention and are not therefore to be considered to be limiting of its scope, the invention will be described and explained with additional specificity and detail through the use of the accompanying drawings in which:

FIG. 1 illustrates an Ethernet link between local and remote link partners;

FIG. 2 shows an example of a controller;

FIG. 3 illustrates one example of a switch;

FIG. 4 illustrates one example of a power management tool;

FIG. 5 shows a flow chart of the process steps of the present invention.

Detailed Description

Various embodiments of the invention are discussed in detail below. However, the specific embodiments discussed should be considered merely illustrative of the invention, and it will be apparent to those skilled in the art that other elements and configurations may be used without departing from the spirit and scope of the invention.

In the context of data or computing centers, dynamic power management is based on an analysis of demand levels. Many conventional solutions are designed to perform power management based on inferences about the level of demand of a data or computing center, which are guesses at best.

In the present invention, the inference of demand levels is replaced with an analysis of power consumption and an analysis of network demand of computing center resources (e.g., servers, switches), which may affect the power saving state in one or more hardware subsystems included in the computing center resources. Such analysis provides a more accurate measure of the actual application of computing services provided by the data or computing center. An example of such analysis is provided by an Energy Efficient Ethernet (EEE) network.

Generally, EEE networks attempt to save power when the network's traffic is not at its maximum capacity. This can be used to minimize performance impact while maximizing energy savings. In a broad sense, the EEE control policy for a particular link in the network determines when to enter a power saving state, what power saving state (e.g., power saving class) to enter, how long the power saving state is maintained, what power saving state to transition from a previous power saving state, etc. The EEE control strategy may make decisions taking into account both static settings established by the IT manager and the link's own traffic attributes.

Fig. 1 illustrates an example of a link used by the EEE control policy, as shown, the link supports communication between a first link partner 110 and a second link partner 120. In different examples, the link partners 110 and 120 may represent switches, routers, endpoints (e.g., servers, clients, VOIP phones, wireless access points, etc.), or the like. It should be understood that the link may operate at standard or non-standard link rates (e.g., 2.5G, 5G, 10G, etc.) as well as future link rates (e.g., 40G, 100G, etc.). Links may also be supported by a variety of port types (e.g., backplane, twisted pair, fiber optic cable, etc.) and by a variety of applications (e.g., high coverage ethernet, EPON, etc.). As shown, link partner 110 includes a physical layer device (PHY)112, a Media Access Control (MAC)114, and a host 116, while link partner 120 includes a PHY 122, a MAC 124, and a host 126.

Fig. 2 illustrates an example of a controller system that is part of a server (e.g., audio-visual (AV) server, High Performance Computing (HPC) server) of a data or computing center. As shown, host system 220 is connected to integrated ethernet controller 210. The ethernet controller 210 further comprises a PHY 211, and the PHY 211 is connected to a MAC 212. In the illustrated example, the MAC 212 is coupled to the PCI Express device 216 through a memory controller 213, the memory controller 213 also coupled to a buffer 214 and a processor 215.

Fig. 3 shows an example of a switch system 300, the switch system 300 may represent a router or any other device that includes multi-port switching functionality. As shown, switch system 300 includes a switch 310, switch 310 supporting one internal port and multiple external ports 0-N via MAC and PHY interfaces. Switch 310 may also be supported by buffer 320 and controller 330.

Referring again to FIG. 1, the hosts 116 and 126 may comprise suitable logic, circuitry and/or code that may enable the five highest functional layers of data packets to be transmitted over the link to be operable and/or functional. Because each layer of the OSI model provides services to an upper interface layer, MAC controllers 114 and 124 can provide the necessary services to hosts 116 and 126 to ensure that packets are properly formatted and transmitted to PHYs 112 and 122. The MAC controllers 114 and 124 may comprise suitable logic, circuitry, and/or code that may enable operability and/or functionality of data link layer (layer 2) processes. The MAC controllers 114 and 124 may be configured to implement an ethernet protocol, such as based on the IEEE802.3 standard. PHYs 112 and 122 may be configured to process physical layer requests including, but not limited to, packetization, data transfer, and serialization/deserialization (SERDES).

In transmission, each layer may add its own header to data transmitted from an interface layer above it. During reception, compatible devices with similar OSI stacks may remove the header as information is passed from lower layers up to higher layers.

Generally, controlling the data rate of the link may enable the link partners 110 and 120 to communicate in a more energy efficient manner. In particular, the sub-rate where the link rate is reduced to the main rate can reduce power and thus save power. In one example, this sub-rate may be a zero rate, which may maximize power savings.

One example of subrate (subrating) is through the use of a subset phy (subset phy) technique. In subset PHY techniques, low link usage periods may be accommodated by transitioning the PHY to a lower link rate supported by a subset of the parent PHY (parent PHY). In one embodiment, the subset PHY technique can operate at a lower or subrate by cutting off a portion of the parent PHY.

Another example of subrates is through the use of Low Power Idle (LPI) technology. In general, LPI relies on turning the active channel silent (silent) when there is no signaling, so that power can be saved when the link is cut off. The refresh signals (refresh signals) may be periodically sent to wake up from the sleep mode.

In general, both subset and LPI techniques involve shutting down or making other changes to a portion of the PHY during periods of low link usage. Similar to in PHY, power savings for higher layers (e.g., MAC, controller subsystem, switch subsystem, etc.) may also be obtained by using various forms of subrates.

Fig. 1 further illustrates that the link partners 110 and 120 also include EEE control policy entities 118 and 128, respectively. Generally, the EEE control policy entities 118 and 128 may be designed to be able to determine when to enter an energy saving state, what energy saving state (i.e., energy saving class) to enter, how long the energy saving state is maintained, to what energy saving state to transition from a previous energy saving state, etc.

In general, the EEE control policy entities 118 and 128 may comprise suitable logic, circuitry, and/or code that may enable establishment and/or execution of EEE control policies for a network in which a link may be located. In various embodiments, EEE control policy entities 118 and 128 may be logic and/or functional blocks as implemented in one or more layers (including PHY, MAC, switch, controller, or other subsystems in a host).

The EEE control policy entities 118 and 128 enable analysis of traffic on the physical link and analysis of data operations and/or processing in the link partners 110 and 120. In this manner, EEE control policy entities 118 and 128 may exchange information from or for one or more layers of the OSI hierarchy to establish and/or implement EEE control policies.

IT should be noted that the EEE control policy may be designed to be decided based on a combination of static settings established by the IT manager (e.g., time of day considerations) and link-self traffic bandwidth attributes. For example, the EEE control policy may be designed to check the idle or non-idle state of a port, queue, buffer, etc. to determine whether to transition out of or into a power saving state.

The EEE control policy may also be designed to make decisions based on dynamic considerations (e.g., type of traffic, identity of the user/source device). Regardless of the particular nature of the EEE control strategy implemented, the result of the EEE control strategy is a decision as to whether to enter an energy saving state, what energy saving state (i.e., energy saving class) to enter, the duration of the energy saving state, etc.

Information from computing center resources supporting EEE functionality may be used to provide actual rather than inferred characteristics of the data or network demand level of the computing center, which is a feature of the present invention. In particular, the link level analysis generated by the computing center resources supporting EEE functionality provides an actual measure of the different resource usage of the data or computing center. This measurement may further be used to infer a level of demand for a plurality of services provided by the data or computing center. This functionality may be illustrated by the high level diagram of FIG. 4, where FIG. 4 shows a plurality of EEE-capable compute farm resources 220 in communication with a power management tool 210₁-220_N。

In one example, compute farm resources 220 supporting EEE functionality₁-220_NMay be designed to communicate power-saving statistics to the power management tool 210. In one embodiment, the power management tool 210 is a power management application running in a management station. In one example, the transmitted power savings statistics may include power savings event data captured for a plurality of power savings events occurring at a particular measurement interval (e.g., N hours). These energy savings statistics provided to the power management tool 210 enable IT administrators to track the actual usage of the compute farm resources as indicated by the energy savings statistics. For example, when some of the compute farm resources that support EEE functionality begin to enter a power-saving state due to low link usage, the power management tool 210 may identify a reduced demand for application services provided by those compute farm resources that support EEE functionality.Then, the IT administrator further considers taking energy saving measures, such as cutting off the server or reusing (reusing) the server.

To further illustrate features of the present invention, reference is now made to FIG. 5. As shown in FIG. 5, the process flow begins at step 502 where the compute farm resource supporting EEE functionality monitors one or more hardware subsystems for an energy savings state. In one example, the hardware subsystem may represent a PHY, which may be designed to enter a power saving state using some form of subrate (e.g., subset PHY, LPI, etc.). It should be appreciated that the compute farm resources supporting EEE functionality may be designed to monitor power savings at different hardware levels (e.g., ports, chips, circuit boards, systems, racks, etc.).

Next, at step 504, the compute farm resource supporting the EEE function reports power-saving statistics based on the monitored event data to the power management tool. It should be understood that the particular type and form of energy conservation statistics reported to the power management tool may depend on the implementation. For example, the reporting frequency of compute farm resources supporting EEE functionality is determined based on the needs of the IT administrator. In one embodiment, SNMP is used to facilitate communication between the power management tool and the compute farm resources that support EEE functionality.

Upon receiving the energy savings statistics from the plurality of compute farm resources supporting EEE functionality, the power management tool determines a utilization level of the data or various services provided by the compute farm in step 506. In one example, the power management tool determines a level of utilization of a particular service based on a percentage of time that one or more compute farm resources supporting EEE functionality are brought into an energy saving state. This utilization level statistic will provide a direct indication of the relative lack of utilization of the compute farm services provided by the one or more compute farm resources. It should be appreciated that the reported energy savings statistics may be used in different ways to identify utilization of a service based on relative activity or inactivity of a particular compute farm resource.

As described above, the power management tool may generate a utilization analysis of different services provided by the computing center based on energy savings statistics provided by the various computing center resources that support EEE functionality. In general, the power management tool may be designed to generate an indication of actual compute farm service utilization based on a set of energy saving statistics indicating network traffic. At step 508, the utilization analysis generated based on these energy savings statistics causes the power management tool to implement control of the compute farm resources that support EEE functionality. For example, the server is switched off or reused as dictated by the service utilization analysis.

Finally, at step 510, the service utilization analysis generated by the power management tool may also be fed back to the various compute farm resources supporting EEE functionality to adjust their own EEE control policies. Indeed, such feedback from the power management tool will enable the various compute farm resources supporting EEE functionality to adjust their power saving decisions in view of the broader service level decisions made using the power management tool. As such, each EEE-capable compute hub resource may implement an EEE control strategy based on a broader subjective initiative than the energy-saving perspective of the respective EEE-capable compute hub resource.

As described above, the present invention allows for more efficient power savings decisions not only from a low level hardware perspective, but also from a high level service perspective. The triggering of high level service power savings is based on consideration of actual activity in the network, however, the triggering of low level hardware power savings may be adjusted based on the more extensive initiatives taken in addition to the individual EEE-capable compute-centric resources.

These and other aspects of the invention will be apparent to those skilled in the art upon review of the foregoing detailed description. While there have been described above many of the features of the present invention, it will be apparent to those of ordinary skill in the art that, upon reading the present disclosure, the invention is capable of other embodiments and of being practiced and carried out in various ways, and thus, the foregoing description should be considered as excluding other embodiments and that the phraseology and terminology employed herein is for the purpose of description and should not be regarded as limiting.

Claims

1. A method for use in a computing center comprising a plurality of server devices for providing application services through the computing center, the method comprising:

a management station in a computing center receiving, from a first server device of the plurality of server devices, first power save information for a first physical layer device of the first server device, the first physical layer device being coupled to a first media access control device in the first server device, the first physical layer device also being coupled to a second physical layer device in a first link partner device, the first power save information being generated by the first server device based on a power save state used by the first physical layer device in response to a low traffic utilization condition detected by the first server device;

the management station receiving, from a second server device of the plurality of server devices, second power saving information of a third physical layer device of the second server device, the third physical layer device being coupled to a second media access control device of the second server device, the third physical layer device also being coupled to a fourth physical layer device of a second link partner device, the second power saving information being generated by the second server device based on a power saving state used by the third physical layer device in response to a low traffic utilization condition detected by the second server device;

the management station deciding a utilization level of an application service provided by the computing center based on the received first energy saving information issued from the first server apparatus and the second energy saving information issued from the second server apparatus; and

based on the decision, one of the server devices for providing the application service is switched off.

2. The method of claim 1, wherein the receiving comprises receiving from a switch.

3. The method of claim 1, wherein the receiving comprises receiving from a server.

4. The method of claim 1, wherein the receiving comprises receiving statistics of physical layer devices entering a power saving state.

5. The method of claim 1, wherein the shutting down comprises shutting down a server.

6. The method of claim 1, wherein the disconnecting comprises reusing a server.

7. A method for use in a computing center, the computing center including a plurality of server devices for providing application services through the computing center, the method comprising:

a management station in a computing center receiving, from a first server device of the plurality of server devices, first power saving information of a first hardware subsystem of the first server device, the first power saving information being generated by the first server device based on a first power saving state used by the first hardware subsystem, so that when no traffic sent by a first physical layer of the first server device to a second physical layer in a first link partner device is monitored, the first physical layer device and the second physical layer device are coupled in response to a low traffic utilization condition detected by the first server device;

the management station receiving, from a second server apparatus of the plurality of server apparatuses, second power saving information of a second hardware subsystem of the second server apparatus, the second power saving information being generated by the second server apparatus based on a second power saving state used by the second hardware subsystem, so that when it is monitored that there is no traffic sent by a third physical layer of the second server apparatus to a fourth physical layer in a second link partner apparatus, the third physical layer apparatus and the fourth physical layer apparatus are coupled in response to a low traffic utilization condition detected by the second server apparatus;

8. A method of managing energy efficient control policies for server devices, wherein the server devices are a plurality of server devices at a computing center, the plurality of server devices being configured to provide application services through the computing center, the method comprising:

based on the determination, changing an energy-efficient control strategy in the first server device, wherein the energy-efficient control strategy controls the first hardware subsystem to enter a power-saving state.

9. The method of claim 8, wherein the monitoring comprises monitoring by a server.

10. The method of claim 8, wherein the monitoring comprises monitoring a physical layer device in the first server device to enter the power saving state.