Optimize server BP. Two 800W or two 450W + PowerSafeguard?
How to calculate what BP is enough to ensure the continuous operation of the server? The answer is not as straightforward as it would seem. Calculate the maximum power consumption of all components and ensure the delivery of the required power from one power supply unit, and then install two such power supplies to the server for redundancy - this is the traditional approach. However, thanks to the PowerSafeguard technology, in most cases, it is possible to do with power supply units with a lower nominal value than is required by the results of such a calculation, and at the same time be sure that the server is up and running.
PowerSafeguard is a technology that automatically regulates server power consumption to maintain continuous operation in the event of a failure of a part of the power supply, if the power of the remaining power supply units is insufficient for a full load. Fujitsu has been offering this technology for some time now in its RX200 S7, RX300 S7, RX350 S7, TX300 S7 servers. All of these servers use two types of standardized power supplies - 450 W or 800 W.
PowerSafeguard allows you to operate the server on 450 W power supplies and ensure continuity of operation when one of the power supply fails, even if the peak power consumption of this configuration could theoretically exceed 450 W (the maximum peak load over the power supply is 30%, i.e. configurations with a theoretical peak power consumption of up to 585 watts will be able to work on a 450 watt power supply.
')
In situations where the actual power consumption of the system is on the verge of the power supply capabilities, the PowerSafeguard technology reduces the p-state (respectively, the clock frequency of the processor cores), ensuring the overall power consumption of the system is within the power of one power supply. The continuity of the server is guaranteed, although in some periods of time will be reduced system performance.
Explain the essence of the technology is convenient by example. Take the server PRIMERGY RX300 S7. Let him have a fairly rich configuration - two Xeon E5-2680 processors, 24x4 = 96GB of memory, 8 SAS drives, 1.2TB PCIe-SSD, RAID 5/6 with FBU, a FibreChannel controller and a BlueRay writer.
Peak power consumption of this system can be checked using the PowerCalculator function in the Fujitsu configurator . We look at the load on the PSU, that is, DC-load.
At maximum server load it will be 524.8 watts. Of the two possible power supply options (450 W and 800 W), there is a desire to put the second one. However, we will deliver 2x 450W. Naturally, two such BPs fully load the load, while working in a balanced mode. What happens if one of them burns out?
It should be noted that the server’s power consumption is not constant, but varies depending on server load (thanks to Intel’s Xeon E5 platform flexibility, including TurboBoost 2.0 technology). The graph of power consumption over time will look like a wavy line.
As the PowerCalculator shows, with a very typical server load level of 70% - it consumes no more than 378 watts!
That is, in most cases, even a richly configured server will continue to work perfectly on one 450-watt power supply. However, where we would have power consumption peaks with two power supplies, PowerSafeguard will work. Through the integrated server management chip (iRMC), the p-state of the processors will be reduced to the minimum, and then will increase until the power consumption “rests” against the power supply limit, or until it turns out that it is not necessary to limit the processor - because reducing the server load this limit will be met without the intervention of the PowerSafeguard technology.
Thus, when operating in the “emergency” mode (on one power supply unit, until the second one is restored), performance will only be observed for certain periods of time. In this case, in the case of a typical server load of about 70%, full performance will be available. Only bursts of the graphics are cut off.
Naturally, for a specific configuration, it is better to check the power consumption parameters at different loads using a calculator. But I would like to note that, for example, a truly “predatory” for power consumption configuration is taken, and most of the most popular server configurations of this type, even in peak situations, consume less than 450 watts. So the use of 450-watt PSU instead of 800-watt is justified. In addition to the price, I will add another argument: the power supply units have the highest energy efficiency with a high load level, therefore, from the energy efficiency point of view, a 450 W power supply unit operating with a load of more than 60% is better than delivering the same power of a 800 W power supply loaded at 34%.
PowerSafeguard is a technology that automatically regulates server power consumption to maintain continuous operation in the event of a failure of a part of the power supply, if the power of the remaining power supply units is insufficient for a full load. Fujitsu has been offering this technology for some time now in its RX200 S7, RX300 S7, RX350 S7, TX300 S7 servers. All of these servers use two types of standardized power supplies - 450 W or 800 W.
PowerSafeguard allows you to operate the server on 450 W power supplies and ensure continuity of operation when one of the power supply fails, even if the peak power consumption of this configuration could theoretically exceed 450 W (the maximum peak load over the power supply is 30%, i.e. configurations with a theoretical peak power consumption of up to 585 watts will be able to work on a 450 watt power supply.
')
In situations where the actual power consumption of the system is on the verge of the power supply capabilities, the PowerSafeguard technology reduces the p-state (respectively, the clock frequency of the processor cores), ensuring the overall power consumption of the system is within the power of one power supply. The continuity of the server is guaranteed, although in some periods of time will be reduced system performance.
Explain the essence of the technology is convenient by example. Take the server PRIMERGY RX300 S7. Let him have a fairly rich configuration - two Xeon E5-2680 processors, 24x4 = 96GB of memory, 8 SAS drives, 1.2TB PCIe-SSD, RAID 5/6 with FBU, a FibreChannel controller and a BlueRay writer.
Peak power consumption of this system can be checked using the PowerCalculator function in the Fujitsu configurator . We look at the load on the PSU, that is, DC-load.
At maximum server load it will be 524.8 watts. Of the two possible power supply options (450 W and 800 W), there is a desire to put the second one. However, we will deliver 2x 450W. Naturally, two such BPs fully load the load, while working in a balanced mode. What happens if one of them burns out?
It should be noted that the server’s power consumption is not constant, but varies depending on server load (thanks to Intel’s Xeon E5 platform flexibility, including TurboBoost 2.0 technology). The graph of power consumption over time will look like a wavy line.
As the PowerCalculator shows, with a very typical server load level of 70% - it consumes no more than 378 watts!
That is, in most cases, even a richly configured server will continue to work perfectly on one 450-watt power supply. However, where we would have power consumption peaks with two power supplies, PowerSafeguard will work. Through the integrated server management chip (iRMC), the p-state of the processors will be reduced to the minimum, and then will increase until the power consumption “rests” against the power supply limit, or until it turns out that it is not necessary to limit the processor - because reducing the server load this limit will be met without the intervention of the PowerSafeguard technology.
Thus, when operating in the “emergency” mode (on one power supply unit, until the second one is restored), performance will only be observed for certain periods of time. In this case, in the case of a typical server load of about 70%, full performance will be available. Only bursts of the graphics are cut off.
Naturally, for a specific configuration, it is better to check the power consumption parameters at different loads using a calculator. But I would like to note that, for example, a truly “predatory” for power consumption configuration is taken, and most of the most popular server configurations of this type, even in peak situations, consume less than 450 watts. So the use of 450-watt PSU instead of 800-watt is justified. In addition to the price, I will add another argument: the power supply units have the highest energy efficiency with a high load level, therefore, from the energy efficiency point of view, a 450 W power supply unit operating with a load of more than 60% is better than delivering the same power of a 800 W power supply loaded at 34%.
Source: https://habr.com/ru/post/162151/
All Articles