When a Windows VPS feels slow, the useful question is not “is it slow” but “slow at what, and since when”. Performance Monitor and Event Viewer answer both questions, are installed by default, and cost nothing. This article shows how to build a diagnostic baseline on a fresh Windows VPS, capture and interpret live counters, and use the event log to turn a vague complaint into a specific, fixable finding.
Everything here uses inbox tooling on Windows Server 2019 or 2022: Performance Monitor (perfmon.msc), the Get-Counter cmdlet, and Get-WinEvent. No agents, no third-party APM, no dashboards to maintain.
Establishing a Baseline Before There Is a Problem
A baseline is the only way to know what “normal” looks like on your workload. A CRM application that sits at 70 percent CPU is healthy for that application; the same number on an idle DNS resolver is alarming. Capture counters when the server is behaving well, at the same time of day you would normally investigate, so that later comparisons are not confounded by batch jobs or backup windows.
Use Get-Counter -Counter "\Processor(_Total)\% Processor Time","\Memory\Available MBytes","\PhysicalDisk(_Total)\Avg. Disk sec/Transfer","\Network Interface(*)\Bytes Total/sec" -SampleInterval 5 -MaxSamples 120 to take a ten-minute sample, then Export-Clixml the result with a timestamped filename. Do this weekly for a month and you have a defensible picture of typical load, along with disk latency that you can quote when a provider asks whether the storage is performing as sold.
Counters Worth Watching and What They Mean
Four counter families cover most VPS performance problems. \Processor(_Total)\% Processor Time plus \System\Processor Queue Length tells you whether CPU is saturated — sustained queue length above 2 per logical core means work is waiting. \Memory\Available MBytes and \Memory\Pages/sec reveal pressure; a server with 4 GB free but heavy paging is worse off than one with 800 MB free and no paging.
For storage, \PhysicalDisk(_Total)\Avg. Disk sec/Read is the single most important counter on a virtual machine: under 10 ms is fine for most workloads, 20-50 ms indicates a constrained backend, and anything above 100 ms is a user-visible stall. For network, \Network Interface(*)\Output Queue Length plus Bytes Total/sec distinguishes a saturated link from an application that simply talks a lot. Add \Paging File(_Total)\% Usage if the guest has a fixed page file on a slow volume.
Collector Sets and Data Collector Logs
Live monitoring in perfmon.msc only records while you are looking at it, which is exactly when the problem is least likely to occur. Create a Data Collector Set instead: expand Data Collector Sets > User Defined, create a set of type Performance counter alert or Performance counter data collector, add the counters above at a 15-second interval, and set a schedule.
The files land as .blg binaries in the output directory and can be loaded back into Performance Monitor or read with Get-Counter -Counter ... -MaxSamples. A simple alert collector with the threshold \PhysicalDisk(_Total)\Avg. Disk sec/Transfer > 0.05 will write an entry to the Application log every time storage latency crosses 50 ms, giving you a timestamped record of every slow period across a week, which is far more useful than a single snapshot taken during a complaint.
Event Viewer: System Log Priorities
The System log carries the hardware and driver signals. Event ID 6008 records an unexpected shutdown with its timestamp, 41 (Kernel-Power) indicates the machine rebooted without a clean shutdown, and 1001 (BugCheck) contains the stop code. Get-WinEvent -FilterHashtable @{LogName='System'; Id=6008,41,1001; StartTime=(Get-Date).AddDays(-14)} gives you a two-week crash history in one call, which is the first thing to check when a provider claims the VPS “rebooted itself”.
Disk and storage errors appear as 7, 11, 51, and 153 from the disk and storport providers — bad block retries and controller resets. Any of these recurring on a virtual disk is worth a support ticket with the exact timestamps and the Get-PhysicalDisk | Select FriendlyName,HealthStatus,OperationalStatus output attached. Also watch 7031 and 7034 for services terminated unexpectedly; a service that crashes every few hours explains periodic latency spikes that no counter will attribute correctly.
Application and IIS Logs for Web Workloads
On IIS hosts, request-level evidence lives outside the normal Event Viewer logs. Get-WinEvent -LogName "Microsoft-Windows-WAS/Operational" shows application pool recycles and worker process failures, and the IIS logs under C:\inetpub\logs\LogFiles\W3SVC1 carry time-taken in the last field — the fastest way to identify an endpoint that is slow rather than a server that is slow.
For .NET application faults, the Application log fills with ASP.NET warnings and .NET Runtime errors (source .NET Runtime, event ID 1026 for unhandled exceptions in managed code). Cross-reference the first occurrence timestamp with your performance data. A correlation between a memory counter climbing and a 1026 every hour is usually an unbounded cache or a static collection holding references, not a hosting problem.
Building a One-Line Health Snapshot
For daily checks, a single command block beats opening four consoles. Combine CPU, memory, disk latency, top processes, and recent critical events into one script and mail or save the output. Get-CimInstance Win32_PerfFormattedData_PerfOS_Processor | Select Name,PercentProcessorTime gives per-core CPU without Get-Counter parsing complexity, and Get-Process | Sort WorkingSet64 -Descending | Select -First 10 names the likely culprit.
Write the output to a dated file so you accumulate history. After two weeks you can answer questions like “was it slow before the last update” by diffing two snapshots, which is exactly the kind of evidence a hosting provider’s support team will ask for. Built-in tooling is not as pretty as a dashboard, but it is always available, even when an agent has crashed precisely when you need it.
Deciding Whether the Problem Is Yours or the Host’s
This is where measurement pays off. If \PhysicalDisk(_Total)\Avg. Disk sec/Read is high while CPU and memory are idle, your application is not the problem — the storage backend is. If CPU is pegged at 100 percent with a long processor queue, look at Get-Process before opening a ticket.
Before contacting support, gather the counters, the crash history from 6008 and 41, and the exact times. Review the performance characteristics a Windows VPS plan is expected to deliver — advertised IOPS and guaranteed CPU share — so you are comparing measurements against commitments rather than against expectations. That single step converts most “it feels slow” tickets into specific, actionable ones.



