vSphere perfQuery returns incorrect counter values(spikes)

I'm writing a perl script calling perfQuery to retrieve the counters. The returned counters have spikes every 8-9 hours. Though if I look at the performance chart reports by vSphere Web Client, I don't see the spikes at all(below). Anyone has idea why this happened and how to read the counters correctly?

Any help is appreciated.

vSphere Web Client performance chart:

Object,Measurement,Rollup,Units,Latest,Maximum,Minimum,Average

vmabc,Usage,Average,%,0.16,0.17,0.16,0.162

My PerfQuerySpec is:

<!-- array start -->

<querySpec>

   <entity type="ManagedObject">vm-168</entity>

   <!-- optional -->

   <startTime>2017-02-24T02:30:00Z</startTime>

   <!-- optional -->

   <endTime>2017-02-24T16:00:00Z</endTime>

   <!-- optional -->

<!-- optional -->

<maxSample>10</maxSample>

   <!-- array start -->

   <!-- optional -->

   <metricId>

      <counterId>2</counterId>

      <instance>*</instance>

   </metricId>

   <!-- array end -->

   <!-- optional -->

   <!-- optional -->

   <format>normal</format>

</querySpec>

<!-- array end -->

queryPerf output: