The present invention relates to a so-called, open air cooled information processing system such as an open air cooled data center comprised of a cluster of information processing devices such as server machines and cooling equipment for cooling these devices and an operation management method of the information processing devices in such system. In particular, the present invention relates to an open air cooled information processing system or a locally coolable information processing system and a method for allocating workloads to information processing devices in a cluster in such system.
Currently, with increase in the amount of heat generated by information processing devices such as server machines which augment exclusively, cooling electricity, i.e., electric power consumed by cooling equipment for cooling the heat generated also increases. A problem encountered is to save overall electricity, a total of both electric power consumed by the server machines and the cooling electricity.
With the aim of electric power saving of a data center, various approaches are made to save electric power in regard to information processing devices such as server machines, cooling equipment, and operation management, respectively.
With respect to the information processing devices, for example, improving performance per wattage consumed through the use of low-power devices and exercising a power-saving function by operating state changeover depending on workload are promoted.
With respect to the cooling equipment, it is contrived to accomplish electric power saving by taking in open air and carrying out cooling by air blow only, unlike a heat pump type which uses a compressor in air conditioning.
In addition, with respect to the operation management, monitoring operative/inoperative status information, job scheduling, and improving the operation efficiency of devices in a cluster and consolidation by means of virtualization, etc. are performed.
Cooling equipment is generally designed based on a total maximum rated power of server machines in a cluster. In practical operation, however, all server machines may not be necessarily used at the same time and the load of a server machine used may not always reach a maximum load.
In future, as operation management technology for data centers will evolve, consolidation of server machines for electric power saving using virtualization technology is foreseen to be utilized. In addition, cooling equipment that takes in open air is foreseen to be used together or locally cooling equipment with a plurality of air conditioning units is foreseen to be used together. In this foreseen situation, uneven distribution of electric power across devices in a cluster may cause a back stream or recirculation of cooling airflow. In consequence, rather adversely, there is a possibility that heat accumulation takes place and results in a fault such as an abnormal intake air temperature of a certain server.
Until now, as operation management methods for activating and deactivating servers depending on their loads, some patent documents which will be outlined below have already been known.
First, according to Japanese Published Unexamined Patent Application No. 2011-40083 mentioned below, operation management of an information processing system comprised of a cluster of information processing devices such as servers, storages, and a network and equipment for supplying power to or cooling these devices is implemented such that allocation of workloads to the devices in the cluster and operation simulation of the equipment are executed in a common virtual space in an integrated manner by reproducing, in the virtual space, the operative/inoperative statuses of the devices in the cluster and the equipment and how the devices and the equipment are deployed in a real space. In particular, this document provides a technique suitable for carrying out integrated power-saving operation management of the information processing devices in the cluster and the equipment.
Then, Japanese Published Unexamined Patent Application No. 2008-171427 mentioned below provides a method for determining an optimal order for reactivating (rebooting) a plurality of servers in a data center. In this method, reactivation is performed in order that is based on an activation order decision policy and it is suggested to provide activation so as to derive proper levels of electric power and wattages.
In addition, in Japanese Published Unexamined Patent Application No. 2011-82799 mentioned below, it is described that, in providing computing resources to a customer system, each data center has knowledge of electric power consumption needed to provide computing resources as electric power information in which an amount of power consumption required for computing resources demanded can be specified and suitable operation of computing resources is implemented by taking such information into account.
As noted above, a feature of the operation management technique according to Japanese Published Unexamined Patent Application No. 2011-40083 above is that allocation of workloads to the devices in the cluster and operation simulation of the equipment are executed in a common virtual space in an integrated manner by reproducing, in the virtual space, the operative/inoperative statuses of the devices in the cluster and the equipment and how the devices and the equipment are deployed. To do this, however, voluminous computation is needed for operation simulation. For example, if a fan for the servers can be put in two operational states, “stop” and “run”, and if there are N servers, simulation runs as many as 2 raised to the Nth power are needed.
In Japanese Published Unexamined Patent Application No. 2008-171427 above, it is stated that reactivation is performed in order that is based on an activation order decision policy, but there is no description about how to determine such policy. Further, in Japanese Published Unexamined Patent Application No. 2011-82799 above, it is stated that a data center has electric power information in which an amount of power consumption required for computing resources demanded can be specified. However, unless such electric power information must be held beforehand as data obtained through numerous case study runs, it is inadequate for actual operation.
In the related art techniques described above, voluminous computation and creating a database by case study runs are need to determine an order in which to activate and deactivate servers. These techniques have a problem in which it takes huge cost to make preparations for putting a system into operation.
Therefore, the present invention has been achieved in view of the problem of the related art techniques described above. More specifically, an object of the present invention is to provide an open air cooled information processing system that is more practical and makes the problem of the related art techniques described above less likely to occur and a load allocation method in such system.
In order to achieve the foregoing object, according to the present invention, there is provided a load allocation method for allocating loads to information processing devices in an open air cooled information processing system that takes in open air and cools a plurality of information processing devices by single air blowing unit. The information processing devices are each positioned across an interface between one space into which open air is taken in and another space into which air after cooling the information processing devices is collected. The load allocation method includes the following steps: when decreasing the number of servers in operation, determining which one of the servers in operation should be deactivated so as to minimize electric power consumed by the air blowing unit by way of numerical analysis performed individually on each of the servers in operation and according to a change in intake air temperature of each server and deactivating the one of the servers; and, when increasing the number of servers in operation, determining which one of the servers out of operation should be activated so as to minimize electric power consumed by the air blowing unit by way of numerical analysis in an inoperative status performed individually on each of the servers out of operation and according to a change in the intake air temperature of each server, and activating the one of the servers.
In another aspect, there is provided a load allocation method in a locally coolable information processing system that cools a plurality of information processing devices by a plurality of air conditioning units adjoining these devices. In the information processing system, the information processing devices are arranged to lie between one space into which cool air supplied by air conditioning units flows in and a space in which air after cooling the information processing devices is collected to the intake side of the air conditioning units adjoining the devices. The method for allocating loads to these information processing devices is performed in the same way as in the foregoing open air cooled air processing system.
In the present invention, in the load allocation method in the open air cooled or locally coolable information processing system, described above, the above numerical analysis maybe performed before start of operation. The load allocation method may preferably include the following steps: setting, as a first server, a server when activated that is determined to bring a result of minimizing the electric power consumed by the air blowing unit and not to make an abnormal change of the intake air temperature of the server, assuming a case where one server has been activated from a state in which all severs are inoperative; setting, as a second server, a server when activated that brings a result of minimizing the electric power consumed by the air blowing unit and does not make an abnormal change of the intake air temperature of the server, assuming a case where one server has further been activated from a state in which the first server is operative; setting, as an n-th server (where n is a natural number smaller than the number of all server), a server when activated that brings a result of minimizing the electric power consumed by the air blowing unit and does not make an abnormal change of the intake air temperature of the server, assuming a case where an (n+1) th server has been activated from a state in which the n-th server is operative; and repeating the step of setting a server as the n-th server until a state in which only one of all servers is operative and storing an order of the servers thus obtained.
According to the present invention and also in order to achieve the foregoing object, there is provided an open air cooled information processing system that takes in open air and cools a plurality of information processing devices by single air blowing unit. The information processing devices are each positioned across an interface between one space into which open air is taken in and another space into which air after cooling the information processing devices is collected. Each of the servers includes means for measuring its intake air temperature. The open air cooled information processing system includes: means for, when decreasing the number of servers in operation, determining which one of the servers in operation should be deactivated so as to minimize electric power consumed by the air blowing unit by way of numerical analysis performed individually on each of the servers in operation and according to a change in the intake air temperature of each server, and deactivating the one of the servers; and means for, when increasing the number of servers in operation, determining which one of the servers out of operation should be activated so as to minimize electric power consumed by the air blowing unit by way of numerical analysis in an inoperative status performed individually on each of the servers out of operation and according to a change in the intake air temperature of each server, and activating the one of the servers. In a similar manner, a locally coolable information processing system is also provided.
According to the present invention described above, it is possible to activate servers in order ensuring that a back stream is always prevented from occurring or has little impact even if occurred. Thus, when a back stream has occurred, the output of the cooling air blower fan required to get rid of it can be minimized. Even if some of server machines are deactivated for electric power saving, it will be possible to avoid that the deactivation results in a fault such as an abnormal intake air temperature of a certain server. Moreover, the present invention is to determine an optimal order in which to activate servers from moment to moment during operation. In a case where consolidation of server machines is utilized for electric power saving and, besides, cooling equipment that takes in open air is used together, uneven distribution of electric power across devices in a cluster may cause a back stream or recirculation of cooling airflow. In consequence, heat accumulation may take place and result in a fault such as an abnormal intake air temperature of a certain server. It will also be possible to avoid this.
In the following, an embodiment of the present invention will be described in detail with reference to the accompanying drawings.
First,
First, cooling airflow produced by taking in open air is described. When open air is taken in (see a flow represented by outline arrows in the figure), impurities (e.g., dust) in the open air are eliminated and the air humidity is conditioned by means for conditioning taken-in open air such as, e.g., a filter, namely, open air conditioning means 11. The air from the open air conditioning means 11 is then sent into a space 12 (so-called cold aisle) onto which the cooling air inlets of the servers are open by a fan 10 that supplies the air as cooling air to a cluster of the above servers 13. Then, the air, after cooling the above servers 13 in the cluster, is collected into a space 14 (so-called hot aisle) onto which the outlets of the servers are open and discharged outside again.
In the open air cooled data center shown, required cooling air is supplied to multiple (N) servers 13 by a single cooling air blower fan 10 (configured to be integral with the open air conditioning means 11). The cooling air inlets of the servers 13 (arranged in two arrays in this example) are open onto the cold aisle in the space of the center. More specifically, this cold aisle 20 is a space defined by ceiling, wall, and floor surfaces in the space of the center, not shown, (excluding the hot aisle defined by aisle capping which will be described later). Cooling air from the above cooling air blower fan 10 (see a flow represented by outline arrows in the figure) passes through a space between the floor 21 and the underfloor base 22 of the center and through openings 22 and is then supplied to the above cold aisle 20 (see the flow represented by outline arrows in the figure)
In the figure, the servers 13 in two arrays are placed with their back sides facing each other and so-called aisle capping 25, which is a partition made of, e.g., transparent plastic sheets or the like, is provided to enclose a space around the back sides of the servers (that is, the space onto which the outlets of the servers are open). By the provision of the aisle capping 25, open air supplied as cooling air into the above cold aisle 20, after cooling the servers 13 in the cluster, is discharged into the space enclosed by the aisle capping 25, namely, the hot aisle 30. Then, the air (hot air) collected into the hot aisle 30 after cooling is circulated to the open air conditioning means 11 through an exhaust duct 40 and discharged to open air again.
That is, the cold aisle 20 and the hot aisle 30 are isolated from each other so that the airs in them are not mixed. Consequently, in the space of the center, because air warmed by heat generated by servers 13 does not increase again the temperature of air that is sucked into other servers 13, the cooling air blower fan 10 is only required to supply a total sum of air volumes demanded by each server 13.
However, according to experiments performed by the present inventors, we have found that, when consolidation of server machines for electric power saving is performed in the foregoing state of the data center, deactivation of a server, when occurs, produces a bypass flow passage through which air can flow back in the server (space inside its housing), because the server's blower fan stops (see a flow represented by a solid arrow in the figure). This poses the following problem. As such a back stream occurs and becomes stronger, a back stream or recirculation of cooling air takes place in the airflow that is flowing until then from the cold aisle 20 toward the hot aisle 30, and eventually heat accumulation takes place. In consequence, to eliminate this problem, it is necessary to increase the output of the cooling air blower fan 10. This adversely results in that electric power required for air conditioning increases to increase the fan output (despite the fact that a server is deactivated).
Thus, in the present invention, to address the foregoing problem, countermeasures are taken as below. Since the intake air temperature of a server rises when a back stream takes place, a temperature sensor (reference numeral 105 in
Now, returning to
For activating a new server, in the open air cooled information processing system (open air cooled data center) configured as described previously, the server load increase/decrease noticing device 101 that gives notice of an increase/decrease in server load gives notice to activate a server out of operation to the server activation/deactivation commanding and executing device 102 that commands and executes activation/deactivation of each server (step S31). Upon receiving this instruction, the server activation/deactivation commanding and executing device 102 refers to the server and air conditioning unit activation/deactivation order storing device 103, and determines whether a server to be activated is determined (step S32). In other words, it determines whether or not one server to be activated from the current status is stored in the storing device 103, and, if a server to be activated is stored (“Yes” in
Otherwise, if the storing device does not have information on a server to be activated, as determined at step S31 above (“No” in
Stated differently, when decreasing the number of servers in operation, the following steps are executed: determining which one of the servers in operation should be deactivated so as to minimize electric power consumed by the air blower (cooling air blower fan 10) by way of numerical analysis performed individually on each of the servers in operation, judging whether or not an abnormal change in the intake air temperature of each server occurs, and deactivating the thus determined server. On the other hand, when increasing the number of servers in operation, the following step is executed: determining which one of the servers out of operation should be activated so as to minimize electric power consumed by the air blowing unit by way of numerical analysis in inoperative status performed individually on each of the servers out of operation and according to a change in the intake air temperature of each server.
The foregoing numerical analysis may be performed before start of operation. In such a case, it is preferable to execute the following steps: setting, as a first server, a server when activated that is determined to bring a result of minimizing the electric power consumed by the above air blowing unit and not to make an abnormal change of the intake air temperature of the server, assuming a case where one server has been activated from a state in which all severs are inoperative; setting, as a second server, a server when activated that brings a result of minimizing the electric power consumed by the above air blowing unit and does not make an abnormal change of the intake air temperature of the server, assuming a case where one server has further been activated from a state in which the first server is operative; setting, as an n-th server (where n is a natural number smaller than the number of all server), a server when activated that brings a result of minimizing the electric power consumed by the above air blowing unit and does not make an abnormal change of the intake air temperature of the server, assuming a case where an (n+1) th server has been activated from a state in which the n-th server is operative; and repeating the step of setting a server as the n-th server until a state in which only one of all servers is operative and storing an order of the servers thus obtained.
Next, descriptions are provided with regard to deactivating a server in operation with reference to
Information stored in the above server and air conditioning unit activation/deactivation order storing device 103 is valid, unless a change is made to the configuration of servers or the like. Next time and subsequently, the operative/inoperative statuses of the servers are notified to the server and air conditioning unit determining unit 104 that collects information on the operative/inoperative statuses and the temperatures of the servers and determines one server to be activated. Then, the server and air conditioning unit determining unit 104 activates servers that are now inoperative one by one through the server activation/deactivation commanding and executing device 102 and acquires and aggregates information on the intake air temperatures of the respective servers from the intake air temperature sensors 105 of the servers. Thereby, it is possible to dispense with the operation of determining one server to be activated, as a server, activation of which brought a result of minimizing a maximum value of the intake air temperatures of the servers in operation.
In the above embodiment, among the devices that execute above-described operations, it has been explained that the server and air conditioning unit determining unit 104 is a device that collects information on the operative/inoperative statuses and the temperatures of the servers and determines one server to be activated. However, the present invention is not so limited. The server and air conditioning unit determining unit 104 for determining one server to be activated may execute simulation discussed in Japanese Published Unexamined Patent Application No. 2011-40083 mentioned previously.
Even in a case where simulation is executed, according to the foregoing present invention, simulation runs as many as 2 raised to the Nth power, as mentioned previously, need not be performed, but the simulation runs can be reduced to N times. More specifically, if, for example, the number of servers is 20, simulation runs as many as 2 raised to the 20th power (=1048576) can be reduced to 20 times.
Using
Next,
Next,
Number | Date | Country | Kind |
---|---|---|---|
2012-051524 | Mar 2012 | JP | national |
2013-038171 | Feb 2013 | JP | national |