# US 10,503,225 B2 · Nonhomogeneous server arrangement > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 10,503,225 B2 - Document ID: US10503225B2 - Status: Granted - Kind: B2 - Application number: 16/172,511 - Filed: 2018-10-26 - Published: 2019-02-28 - Granted: 2019-12-10 - Earliest claimed: 2013-12-31 - Assignees: Microsoft Technology Licensing, LLC - Applicants: Microsoft Technology Licensing, LLC - Inventors: Euan Peter Garden; John Raymond Justice - Web page: https://patentgrep.com/patent/US10503225B2 ## CPC - G06F 1/206: {comprising thermal management} - G06F 1/3206: Monitoring of events, devices or parameters that trigger a change in power modality - G06F 9/5094: {where the allocation takes into account power or heat criteria (power management in computers in general G06F1/3203; thermal management in computers in general G06F1/206)} - Y02D 10/00: Energy efficient computing, e.g. low power processors, power management or thermal management - G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N) - Y02D: CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE ## Abstract Aspects of the present invention describe a nonhomogeneous server deployment in which different classes of servers are placed within a data center unit, such as a rack or chassis. In one aspect, the cooling capacity for the unit is intentionally sized to be incapable of providing enough cooling to maintain an acceptable operational temperature for the servers, if all servers in the rack are simultaneously in an active processing mode. The control fabric maintains an acceptable operating temperature within the unit by assigning workloads to only a portion of the servers within the unit at a given point in time. In one aspect, servers are arranged within a data center unit according to anticipated peak periods of usage. Units can be arranged to be adjacent to servers optimized for a specialized workload having a peak period of usage that differs from each other. ## Claims ### Claim 1 A data center having nonhomogeneous servers interleaved within a rack comprising: a first quantity of graphics optimized servers having a first hardware configuration and a second quantity of general processing optimized servers having a second hardware configuration that is different from the first hardware configuration, the first quantity of graphics optimized servers and the second quantity of general processing optimized servers being power balanced to use a substantially equal amount of power at peak power, and an amount of cooling available being insufficient to cool both the first quantity of graphics optimized servers and the second quantity of general processing optimized servers in active processing mode, wherein the first hardware configuration comprises a graphics processing unit (“GPU”) and the second hardware configuration does not include GPU, and wherein interleaved within a rack means that the first quantity of graphics optimized servers and the second quantity of general processing optimized servers are arranged within the rack in an alternating arrangement. ### Claim 2 (depends on 1) The data center of claim 1, wherein the graphics optimized server outputs a rendered video game image over a wide area network to a remotely located gaming device. ### Claim 3 (depends on 1) The data center of claim 1, wherein the graphics optimized server has a central processing unit (“CPU”), and a video encoder, and wherein a maximum power usage of the GPU comprises more than 40% of the graphics optimized server's maximum power usage. ### Claim 4 (depends on 1) The data center of claim 1, wherein the first class of server and the second class of server generate a substantially equal amount of heat when in use, and the amount of available cooling for the data center unit is less than 60% of an amount needed to adequately cool all servers in the first class of server and the second class of server running in the active processing mode. ### Claim 5 (depends on 1) The data center of claim 1, wherein the first quantity and the second quantity are substantially equal. ### Claim 6 (depends on 1) The data center of claim 1, wherein the first quantity of graphics optimized servers is designed for a workload with a peak usage during a first time period that does not overlap with a second time period for which the second quantity of general processing optimized servers is designed. ### Claim 7 (depends on 1) The data center of claim 1, wherein the rack uses vertical cooling provided by one or more fans located on top of the rack or underneath the rack, and wherein the amount of cooling available does not exceed 70% of what is adequate to facilitate simultaneous operation of the first quantity of graphics optimized servers and the second quantity of general processing optimized servers in the active processing mode. ### Claim 8 A method for managing workloads within a data center, the method comprising: during a first time period, setting substantially all of a first class of server within a data center rack to a low power mode, the data center rack having a nonhomogeneous deployment of servers comprising at least the first class of server and a second class of server, an amount of available cooling for the data center rack being insufficient to cool all servers in the first class of server and the second class of server running in an active processing mode, wherein the first class of server has a first hardware configuration and the second class of server has a second hardware configuration, wherein the first hardware configuration comprises a graphics processing unit (“GPU”) and the second hardware configuration does not include GPU, wherein the nonhomogeneous deployment is in a repeating pattern of a unit of the first class of server adjacent to a unit of the second class of server; and during a second time period in the data center, setting a majority of the second class of server within the data center rack to the low power mode, the second time period not substantially overlapping with the first time period. ### Claim 9 (depends on 8) The method of claim 8, wherein the rack uses vertical cooling provided by one or more fans located on top of the rack or underneath the rack, and wherein the amount of cooling available does not exceed 70% of what is adequate to facilitate simultaneous operation of the first quantity of graphics optimized servers and the second quantity of general processing optimized servers in the active processing mode. ### Claim 10 (depends on 8) The method of claim 8, wherein the rack include the same amount of the first class of server and the second class of server. ### Claim 11 (depends on 8) The method of claim 8, wherein the first class of server is a game optimized server and the second class of server is a general purpose server. ### Claim 12 (depends on 8) The method of claim 8, wherein the first class of server and the second class of server generate a substantially equal amount of heat when in use, and the amount of available cooling for the data center unit is less than 60% of an amount needed to adequately cool all servers in the first class of server and the second class of server running in the active processing mode. ### Claim 13 (depends on 8) The method of claim 8, wherein the first class of servers is designed for a workload with a peak usage during a first time period that does not overlap with a second time period during which the second class of servers is designed. ### Claim 14 (depends on 8) The method of claim 8, wherein the first class of server outputs a rendered video game image over a wide area network to a remotely located graphics device. ### Claim 15 A data center system comprising: a data center rack having a total quantity of servers comprising at least graphics optimized servers and general purpose servers, wherein the general purpose servers and the graphics optimized servers are arranged within the rack in an alternating pattern of a graphics optimized server adjacent to a general purpose server; and a data center controller to control an operation of the graphics optimized servers and the general purpose servers within the rack to ensure that at least 40% of the total quantity of servers in the data center rack are in a low power mode and less than 60% of the total quantity of servers are in an active processing mode, an amount of cooling available for the graphics optimized servers and the general purpose servers being insufficient to cool the total quantity of servers in an active processing mode wherein the graphics optimized server outputs a rendered video game image over a wide area network to a remotely located gaming device. ### Claim 16 (depends on 15) The data center system of claim 15, wherein the general purpose servers and the graphics optimized servers output a substantially equal amount of heat when running in the active processing mode. ### Claim 17 (depends on 15) The data center system of claim 15, further comprising a cooling system for the total quantity of servers, the cooling system having a cooling capacity that is not adequate to maintain operational temperatures within the total quantity of servers when more than 70% of the total quantity of servers is in the active processing mode. ### Claim 18 (depends on 17) The data center system of claim 17, wherein the cooling system is a vertical cooling system provided by one or more fans located on top of the rack or underneath the rack. ### Claim 19 (depends on 15) The data center system of claim 15, wherein the graphics optimized servers are designed for a workload with a peak usage during a first time period that does not overlap with a second time period for which the general processing optimized servers are designed. ### Claim 20 (depends on 15) The data center system of claim 15, wherein the graphics optimized servers have a graphics processing unit (“GPU”), a central processing unit (“CPU”), and a video encoder, and wherein a maximum power usage of the GPU comprises more than 40% of the graphics optimized servers' maximum power usage.