Cobbled together from various source, this list represents a best first attempt at a universally applicable list of key Windows metrics regardless of a server's stated purpose.
My plan for "Part 2" involves a similar collection of metrics specific to IIS and ASP.NET applications. Those waters are considerably harder to navigate for me as a non-developer, but they're no less critical to answering questions about concurrent client connections, number of unique requests to an application, tracking memory leaks or other weirdness in worker processes, etc. I'll post that once I have it.
Without further ado:
Memory
| Counter | Description | Use Notes |
| Memory\Available Mbytes | Available system memory in megabytes (MB) |
<10% considered low <5% considered critically low 10MB negative delta per hour indicates likely memory leak |
| Memory\Committed | Amount of committed virtual memory. Also called the "commit charge" in TaskManager |
|
| Memory\System Cache Resident Bytes | Amount of memory consumed by the system file cache. |
Shows as "Metafile" in a memory explorer. In 64-bit systems, memory addressing allows system file cache to consume almost all physical RAM if unchecked. |
| Memory\Pages Input/sec | Rate of total number of pages read from disk to resolve hard page faults |
>10 considered high Compare to Memory\Page Reads/sec to determine number of pages read into memory for each read operation. Will be => Page Reads/sec, and large delta between them might indicate need for more RAM or smaller disk cache |
| Memory\Pages/sec | Rate of total number of pages read from and written to disk to resolve hard page faults |
>1000 considered moderate, as memory may be
getting low >2000 considered critical, as system likely experiencing delays due to high reliance on slow disk resources for paging Sum of Pages Input/sec and Pages Output/sec |
| Memory\Page Faults | Total number of hard and soft page faults | Pages/sec can be calculated as a % of Page Faults to determine what % of total faults are hard faults |
| Process(*)\Handle Count | Total count of concurrent handles in use by the specified process. |
As this number constantly fluctuates, the delta between high and
low values is most important. Max - Min = !> 1000 Consistent large number or aggressive upward trend of handle count commonly causes memory leaks |
Processor
| Counter | Description | Use Notes |
| System\Processor Queue Length | Total number of queued threads waiting to be processed for all processors. |
=>10 considered high for multi-processor
system PQL/n, where n = number of logical processors, gives per-core queue length. If % Processor Time is high (=>90%) and per-core PQL is =>2, there is a performance bottleneck. It is not uncommon to experience low % Processor Time and a per-core PQL => 2 depending on efficiency of the requesting application's threading logic |
| Processor\%Processor Time | Percentage of time the specified CPU is executing non-idle threads. |
>75% considered moderate and should be
closely monitored >90% considered high, may begin to cause delays in performance >95%-100% considered critically high and will cause major delays in performance This value should be tracked per logical processor. If % Processor Time is high (=>75%) while disk and network utilization is low, consider upgrading or adding processors. |
| Processor\%Interrupt Time | Percentage of time the specified CPU is receiving and servicing hardware interrupts from network adapters, hard disks, and other system hardware | Interrupt rates 30%-50% or higher may indicate a driver or hardware problem |
Disk I/O
| Counter | Description | Use Notes |
| PhysicalDisk\%Disk Time | Percentage of time the selected disk spends servicing read or write requests. | If this value is high relative to nominal CPU and network utilization figures, it is likely disk performance is a problem. |
| PhysicalDisk\Disk Writes/sec | Average number of disk writes per second | Used in conjunction with Disk Reads/sec, general indicator of disk I/O activity |
| LogicalDisk(*)\Avg Disk sec/Writes | Average time in seconds a specified disk takes to process a read request |
>15ms considered slow and worth close
evaluation >25ms considered very slow and likely to negatively impact system performance |
| PhysicalDisk\Avg Write Queue Length | Average number of write requests waiting to be processed |
Used in conjunction with Avg Read Queue Length, this gives an
idea of disk access latency. AWQL/n <= 4, where n is the number of disks in RAID. |
| PhysicalDisk\Disk Reads/sec | Average number of disk reads per second | Used in conjunction with Disk Writes/sec, general indicator of disk I/O activity |
| LogicalDisk(*)\Avg Disk sec/Reads | Average time in seconds a specified disk takes to process a write request |
>15ms considered slow and worth close
evaluation >25ms considered very slow and likely to negatively impact system performance |
| PhysicalDisk\Avg Read Queue Length | Average number of read requests waiting to be processed |
Used in conjunction with Avg Write Queue Length, this gives an
idea of disk access latency. ARQL/n <= 4, where n is the number of disks in use. |
Network I/O
| Counter | Description | Use Notes |
| Network Interface(*)\Total Bytes/sec | Measure of total bytes sent and received per second for the specified network adapter | If >50% of Current Bandwidth value under typical load, problems during peak times are likely. |
| Network Interface(*)\Current Bandwidth |
Estimate of the current bandwidth in bits per second (bps) available to the specified NIC.
Considered "nominal bandwidth" where accurate estimation impossible or where bandwidth doesn't vary |
To estimate current NIC utilization, use the following formula:
Nic Utilization = ((Max Bytes Total/Sec * 8) / (Current Bandwith)) * 100 >30% NIC utilization on shared network considered high |
| Network Interface(*)\Output Queue Length | Total number of threads waiting for outbound processing by the specified NIC |
>1 sustained considered
high >2 sustained considered critically high |
