RHEL 8

This page describes the Optimization Pack for the component type RHEL 8.

Metrics

CPU

Metric	Unit	Description
cpu_num	CPUs	The number of CPUs available in the system (physical and logical)
cpu_util	percent	The average CPU utilization % across all the CPUs (i.e., how much time on average the CPUs are busy doing work)
cpu_util_details	percent	The average CPU utilization % broken down by usage type and cpu number (e.g., cp1 user, cp2 system, cp3 soft-irq)
cpu_load_avg	tasks	The system load average (i.e., the number of active tasks in the system)

Memory

Metric	Unit	Description
mem_util	percent	The memory utilization % (i.e, the % of memory used)
mem_util_nocache	percent	The memory utilization % (i.e., the % of memory used) without considering memory reserved for caching purposes
mem_util_details	percent	The memory utilization % (i.e., the % of memory used) broken down by usage type (e.g., active memory)
mem_used	bytes	The total amount of memory used
mem_used_nocache	bytes	The total amount of memory used without considering memory reserved for caching purposes
mem_total	bytes	The total amount of installed memory
mem_fault_minor	faults/s	The number of minor memory faults (i.e., faults that do not cause disk access) per second
mem_fault_major	faults/s	The number of major memory faults (i.e., faults that cause disk access) per second
mem_fault	faults/s	The number of memory faults (major + minor)
mem_swapins	pages/s	The number of memory pages swapped in per second
mem_swapouts	pages/s	The number of memory pages swapped out per second

Network

Metric	Unit	Description
network_tcp_retrans	retrans/s	The number of network TCP retransmissions per second
network_in_bytes_details	bytes/s	The number of inbound network packets in bytes per second broken down by network device (e.g., wlp4s0)
network_out_bytes_details	bytes/s	The number of outbound network packets in bytes per second broken down by network device (e.g., eth01)

Disk

Notice: you can use a device custom filter to monitor a specific disk with Prometheus. You can find more information on Prometheus queries and the %FILTERS% placeholder here: Prometheus provider and here: Prometheus provider metrics mapping.

Metric	Unit	Description
disk_swap_util	percent	The average space utilization % of swap disks
disk_swap_used	bytes	The total amount of space used by swap disks
disk_util_details	percent	The utilization % of disk, i.e how much time a disk is busy doing work broken down by disk (e.g., disk D://)
disk_iops_writes	ops/s	The average number of IO disk-write operations per second across all disks
disk_iops_reads	ops/s	The average number of IO disk-read operations per second across all disks
disk_iops	ops/s	The average number of IO disk operations per second across all disks
disk_response_time_read	seconds	The average response time of IO read-disk operations
disk_response_time_worst	seconds	The average response time of IO disk operations of the slowest disk
disk_response_time_write	seconds	The average response time of IO write-disk operations
disk_response_time_details	ops/s	The average response time of IO disk operations broken down by disk (e.g., disk /dev/nvme01 )
disk_iops_details	ops/s	The number of IO disk-write operations of per second broken down by disk (e.g., disk /dev/nvme01)
disk_io_inflight_details	ops	The number of IO disk operations in progress (outstanding) broken down by disk (e.g., disk /dev/nvme01)
disk_write_bytes	bytes/s	The number of bytes per second written across all disks
disk_read_bytes	bytes/s	The number of bytes per second read across all disks
disk_read_write_bytes	bytes/s	The number of bytes per second read and written across all disks
disk_write_bytes_details	bytes/s	The number of bytes per second written from the disks broken down by disk and type of operation (e.g., disk /dev/nvme01 and operation WRITE)
disk_read_bytes_details	bytes/s	The number of bytes per second read from the disks broken down by disk and type of operation (e.g., disk /dev/nvme01 and operation READ)

Filesystem

Metric	Unit	Description
filesystem_util	percent	The space utilization % of filesystems broken down by type and device (e.g., filesystem of type overlayfs on device /dev/loop1)
filesystem_used	bytes	The amount of space used on the filesystems broken down by type and device (e.g., filesystem of type zfs on device /dev/nvme01)
filesystem_size	bytes	The size of filesystems broken down by type and device (e.g., filesystem of type ext4 for device /dev/nvme01)

Other metrics

Metric	Unit	Description
proc_blocked	processes	The number of processes blocked (e.g, for IO or swapping reasons)
os_context_switch	switches/s	The number of context switches per second

Parameters

CPU

Parameter	Default Value	Domain	Description
os_cpuSchedMinGranularity	2250000 ns	300000→30000000 ns	Minimal preemption granularity (in nanoseconds) for CPU bound tasks
os_cpuSchedWakeupGranularity	3000000 ns	400000→40000000 ns	Scheduler Wakeup Granularity (in nanoseconds)
os_CPUSchedMigrationCost	500000 ns	100000→5000000 ns	Amount of time (in nanoseconds) after the last execution that a task is considered to be "cache hot" in migration decisions. A "hot" task is less likely to be migrated to another CPU, so increasing this variable reduces task migrations
os_CPUSchedChildRunsFirst	0	0→1	A freshly forked child runs before the parent continues execution
os_CPUSchedLatency	18000000 ns	2400000→240000000 ns	Targeted preemption latency (in nanoseconds) for CPU bound tasks
os_CPUSchedAutogroupEnabled	1	0→1	Enables the Linux task auto-grouping feature, where the kernel assigns related tasks to groups and schedules them together on CPUs to achieve higher performance for some workloads
os_CPUSchedNrMigrate	32	3→320	Scheduler NR Migrate

Memory

Parameter	Default Value	Domain	Description
os_MemorySwappiness	30	0→100	Memory Swappiness
os_MemoryVmVfsCachePressure	100 %	10→100 %	VFS Cache Pressure
os_MemoryVmMinFree	67584 KB	10240→1024000 KB	Minimum Free Memory
os_MemoryVmDirtyRatio	30 %	1→99 %	When the dirty memory pages exceed this percentage of the total memory, processes are forced to write dirty buffers during their time slice instead of continuing to write
os_MemoryVmDirtyBackgroundRatio	10 %	1→99 %	When the dirty memory pages exceed this percentage of the total memory, the kernel begins to write them asynchronously in the background
os_MemoryTransparentHugepageEnabled	never	always never madvise	Transparent Hugepage Enablement
os_MemoryTransparentHugepageDefrag	always	always never madvise defer defer+madvise	Transparent Hugepage Enablement Defrag
os_MemorySwap	swapon	swapon swapoff	Memory Swap
os_MemoryVmDirtyExpire	3000 centisecs	300→30000 centisecs	Memory Dirty Expiration Time
os_MemoryVmDirtyWriteback	500 centisecs	50→5000 centisecs	Memory Dirty Writeback

Network

Parameter	Default value	Domain	Description
os_NetworkNetCoreSomaxconn	128 connections	12→1200 connections	Network Max Connections
os_NetworkNetCoreNetdevMaxBacklog	1000 packets	100→10000 packets	Network Max Backlog
os_NetworkNetIpv4TcpMaxSynBacklog	512 packets	52→15120 packets	Network IPV4 Max Sync Backlog
os_NetworkNetCoreNetdevBudget	300 packets	30→3000 packets	Network Budget
os_NetworkNetCoreRmemMax	212992 bytes	21299→2129920 bytes	Maximum network receive buffer size that applications can request
os_NetworkNetCoreWmemMax	21299→2129920 bytes	21299→2129920 bytes	Maximum network transmit buffer size that applications can request
os_NetworkNetIpv4TcpSlowStartAfterIdle	1	0→1	Network Slow Start After Idle Flag
os_NetworkNetIpv4TcpFinTimeout	60	6 →600 seconds	Network TCP timeout
os_NetworkRfs	0	0→131072	If enabled increases datacache hitrate by steering kernel processing of packets to the CPU where the application thread consuming the packet is running

Storage

Parameter	Default value	Domain	Description
os_StorageReadAhead	128 KB	0→1024 KB	Read-ahead speeds up file access by pre-fetching data and loading it into the page cache so that it can be available earlier in memory instead of from disk
os_StorageNrRequests	1000 packets	100→10000 packets	Network Max Backlog
os_StorageRqAffinity	1	1→2	Storage Requests Affinity
os_StorageQueueScheduler	none	none kyber mq-deadline bfq	Storage Queue Scheduler Type
os_StorageNomerges	0	0→2	Enables the user to disable the lookup logic involved with IO merging requests in the block layer. By default (0) all merges are enabled. With 1 only simple one-hit merges will be tried. With 2 no merge algorithms will be tried
os_StorageMaxSectorsKb	256 KB	32→256 KB	The largest IO size that the OS c