HPC infrastructure measures dataset - turbostat table [11/13] part 1/6
收藏资源简介:
HPC infrastructure measures dataset - turbostat table This record contains the part 1/6 of the turbostat table of the HPC infrastructure measures dataset. Column descriptions: Topology and identification fields: timestamp: The time at which the turbostat measurement was collected. core: The physical processor core identifier. cpu: The logical CPU identifier used by the operating system. die: The processor die identifier. host: The hostname of the machine from which the metrics were collected. l3: The identifier of the L3 cache domain associated with the CPU or core. node: The NUMA node identifier. package: The physical processor package or socket identifier. x2apic: The extended APIC identifier of the logical CPU. Performance and power metrics: core_power_watt: The power consumed by the CPU core domain, expressed in watts. package_power_watt: The power consumed by the whole CPU package, expressed in watts. ram_power_watt: The power consumed by DRAM, expressed in watts. system_power_watt: The total system power consumption, expressed in watts, when provided by the platform. tsc_frequency_mhz: The average Time Stamp Counter (TSC) frequency during the interval, expressed in MHz. average_frequency_mhz: The average CPU frequency over the whole interval, including idle time, expressed in MHz. busy_frequency_mhz: The average CPU frequency while the CPU was busy, expressed in MHz. uncore_frequency_mhz: The uncore frequency, expressed in MHz. busy_percent: The percentage of the interval during which the CPU was actively executing instructions. ipc: The average instructions per cycle (IPC) while the CPU was active. irq: The number of interrupts serviced by the CPU during the sampling interval. nmi: The number of non-maskable interrupts handled during the sampling interval. smi: The number of System Management Interrupts serviced during the sampling interval. usec: The duration of the sample, expressed in microseconds. core_temperature_celsius: The temperature of the CPU core, expressed in degrees Celsius. package_temperature_celsius: The temperature of the CPU package, expressed in degrees Celsius. core_throttle: An indicator of core thermal throttling events during the sampling interval. package_percent: The percentage of the interval during which package-level RAPL throttling was active. ram_percent: The percentage of the interval during which DRAM-level RAPL throttling was active. poll: The number of times the operating system requested the POLL idle state during the sampling interval. poll_minus: The number of times the system requested the POLL idle state when a shallower idle state would have been more appropriate. poll_percent: The percentage of the interval spent in the POLL idle state as requested by software. C-state residency and transition counts: c1: The number of times the operating system requested the C1 idle state during the sampling interval. c1_minus: The number of times the system requested the C1 idle state when a shallower idle state would have been more appropriate. c1_plus: The number of times the system requested the C1 idle state when a deeper idle state would have been more appropriate. c1_percent: The percentage of the interval spent in the C1 idle state as requested by software. cpu_percent_c1: The percentage residency in the hardware C1 core idle state. c1e: The number of times the operating system requested the C1E idle state during the sampling interval. c1e_minus: The number of times the system requested the C1E idle state when a shallower idle state would have been more appropriate. c1e_plus: The number of times the system requested the C1E idle state when a deeper idle state would have been more appropriate. c1e_percent: The percentage of the interval spent in the C1E idle state as requested by software. c2: The number of times the operating system requested the C2 idle state during the sampling interval. c2_minus: The number of times the system requested the C2 idle state when a shallower idle state would have been more appropriate. c2_plus: The number of times the system requested the C2 idle state when a deeper idle state would have been more appropriate. c2_percent: The percentage of the interval spent in the C2 idle state as requested by software. package_percent_pc2: The percentage residency in the hardware package PC2 idle state. c6: The number of times the operating system requested the C6 idle state during the sampling interval. c6_minus: The number of times the system requested the C6 idle state when a shallower idle state would have been more appropriate. c6_plus: The number of times the system requested the C6 idle state when a deeper idle state would have been more appropriate. c6_percent: The percentage of the interval spent in the C6 idle state as requested by software. cpu_percent_c6: The percentage residency in the hardware C6 core idle state. package_percent_pc6: The percentage residency in the hardware package PC6 idle state.
HPC基础设施度量数据集 - turbostat表 本记录为HPC基础设施度量数据集的turbostat表的第1/6部分。 列描述: ### 拓扑与标识字段 timestamp:采集turbostat度量数据的时间戳。 core:物理处理器核心标识符。 cpu:操作系统使用的逻辑CPU标识符。 die:处理器裸片标识符。 host:采集该度量指标的机器的主机名。 l3:与该CPU或核心关联的L3缓存域标识符。 node:NUMA节点(NUMA node)标识符。 package:物理处理器封装或插槽标识符。 x2apic:逻辑CPU的x2apic标识符。 ### 性能与功耗度量 core_power_watt:CPU核心域消耗的功率,单位为瓦特。 package_power_watt:整个CPU封装消耗的功率,单位为瓦特。 ram_power_watt:动态随机存取存储器(DRAM)消耗的功率,单位为瓦特。 system_power_watt:平台提供的系统总功耗,单位为瓦特。 tsc_frequency_mhz:采样区间内的平均时间戳计数器(Time Stamp Counter, TSC)频率,单位为兆赫兹(MHz)。 average_frequency_mhz:整个采样区间(包括空闲时段)内的平均CPU频率,单位为兆赫兹(MHz)。 busy_frequency_mhz:CPU处于忙碌状态时的平均CPU频率,单位为兆赫兹(MHz)。 uncore_frequency_mhz:非核心(uncore)频率,单位为兆赫兹(MHz)。 busy_percent:CPU主动执行指令的时段占总采样区间的百分比。 ipc:CPU活跃状态下的平均每周期指令数(instructions per cycle, IPC)。 irq:采样区间内CPU服务的中断次数。 nmi:采样区间内CPU处理的不可屏蔽中断(non-maskable interrupt, NMI)次数。 smi:采样区间内CPU服务的系统管理中断(System Management Interrupt, SMI)次数。 usec:采样时长,单位为微秒。 core_temperature_celsius:CPU核心温度,单位为摄氏度。 package_temperature_celsius:CPU封装温度,单位为摄氏度。 core_throttle:采样区间内核心热节流事件的指示项。 package_percent:封装级运行平均功率限制(RAPL)节流活动占总采样区间的百分比。 ram_percent:DRAM级RAPL节流活动占总采样区间的百分比。 poll:采样区间内操作系统请求POLL空闲状态的次数。 poll_minus:当存在更浅的空闲状态更为合适时,系统请求POLL空闲状态的次数。 poll_percent:软件请求的POLL空闲状态占用总采样区间的百分比。 ### C-state驻留与转换计数 c1:采样区间内操作系统请求C1空闲状态的次数。 c1_minus:当存在更浅的空闲状态更为合适时,系统请求C1空闲状态的次数。 c1_plus:当存在更深的空闲状态更为合适时,系统请求C1空闲状态的次数。 c1_percent:软件请求的C1空闲状态占用总采样区间的百分比。 cpu_percent_c1:硬件C1核心空闲状态的驻留时间占比。 c1e:采样区间内操作系统请求C1E空闲状态的次数。 c1e_minus:当存在更浅的空闲状态更为合适时,系统请求C1E空闲状态的次数。 c1e_plus:当存在更深的空闲状态更为合适时,系统请求C1E空闲状态的次数。 c1e_percent:软件请求的C1E空闲状态占用总采样区间的百分比。 c2:采样区间内操作系统请求C2空闲状态的次数。 c2_minus:当存在更浅的空闲状态更为合适时,系统请求C2空闲状态的次数。 c2_plus:当存在更深的空闲状态更为合适时,系统请求C2空闲状态的次数。 c2_percent:软件请求的C2空闲状态占用总采样区间的百分比。 package_percent_pc2:硬件封装PC2空闲状态的驻留时间占比。 c6:采样区间内操作系统请求C6空闲状态的次数。 c6_minus:当存在更浅的空闲状态更为合适时,系统请求C6空闲状态的次数。 c6_plus:当存在更深的空闲状态更为合适时,系统请求C6空闲状态的次数。 c6_percent:软件请求的C6空闲状态占用总采样区间的百分比。 cpu_percent_c6:硬件C6核心空闲状态的驻留时间占比。 package_percent_pc6:硬件封装PC6空闲状态的驻留时间占比。



