In Dell’s 13
th generation Dell PowerEdge servers, Intel
ME provides support for Compute Usage Per Second (CUPS) functionality.
The CUPS functionality provides real-time monitoring of CPU, memory
and I/O utilization and system-level utilization index for the system.
Since it is done by Intel ME, it is independent of the OS and does
not consume CPU resources. The Intel ME has a system CUPS sensor which
provides the Computation, Memory and I/O resource utilization value
as CUPS Index. iDRAC monitors this CUPS index for the overall system
utilization and also monitors the instantaneous value of CPU, Memory,
and I/O utilization index.
Utilization information of system resources is obtained by querying
the data from a set of dedicated counters provided by the CPU and
chipset. These counters are called Resource Monitoring Counters or
RMCs. These counters are aggregated by the node manager to measure
the cumulative utilization of each of these system resources that
is read from the iDRAC using existing inter-communication mechanisms
to provide these vital data through out-of-band management interfaces.
Since Intel sensor representation of performance parameters and
index values are for complete physical system, performance data representation
on the interfaces is for the complete physical system, even if the
system is virtualized and hosting multiple virtual hosts.
To display the performance parameters, the supported sensors must
be present in the server.
The four system utilization parameters are:
-
CPU Utilization — There are individual Resource Monitoring
counters (RMCs) for each CPU core which is aggregated to provide cumulative
utilization of all the cores in the system. This utilization is based
on time spent in active state and time spent in inactive state. Each
sample of RMC is taken every six seconds.
-
Memory Utilization — There are individual counters (RMCs)
to measure memory traffic occurring at each memory channel or memory
controller instance. These counters are aggregated to measure the
cumulative memory traffic across all the memory channels on the system. This
is a measure of memory bandwidth consumption and not amount of memory
utilization. iDRAC aggregates it for one minute of period, so it may
or may not match the memory utilization shown by other OS tools such
as top in Linux. Memory bandwidth utilization shown by iDRAC is indication
of whether workload is memory intensive or not.
-
I/O Utilization — There are individual Resource Monitoring
Counters (RMCs), one per root port in the PCI Express Root Complex
to measure PCI Express traffic emanating from or directed to that
root port and the lower segment. These counters are then aggregated
to measure PCI express traffic for all PCI Express segments emanating
from the package. This is measure of IO bandwidth utilization for
the system.
-
System Level CUPS Index — The CUPS index is calculated
by aggregating CPU, Memory, and I/O index considering a pre-defined
load factor of each system resource. The load factor depends on the
nature of the workload run on the system. Thus at any given time,
CUPS Index represents the measurement of the compute headroom available
on the server. Hence, if the system has a large CUPS Index, then there
is limited headroom to place additional workload on that system. As
the resource consumption decreases, the system’s CUPS Index decreases. A
low CUPS Index indicates that there is a large amount of compute headroom
and the server is a main target for receiving new workloads or having
the workload migrated, and the server being placed into a lower power
state in order to reduce power consumption. Such workload monitoring
can then be applied throughout the data center to provide a high-level
and holistic view of the data center’s workload, providing a dynamic
data center solution.
-
NOTE: The CPU,
memory, and I/O utilization indexes are aggregated over one minute.
Therefore, if there are any instantaneous spikes in these indexes,
they may be suppressed. They are indication of workload patterns not
the amount of resource utilization.
The IPMI, SEL, and SNMP traps are generated if the thresholds of
the utilization indexes are reached and the sensor events are enabled.
The sensor event flags are disabled by default. It can be enabled
using the standard IPMI interface.
The required privileges are:
- Login privilege is required to monitor performance data.
- Configure privilege is required for setting warning thresholds
and reset historical peaks.
- Login privilege and Enterprise license is required for reading
historical statics data.