* calling thread wait for that time.
   *
   * The simplest way to maintain a rate of QPS is to keep the timestamp of the last granted
   * request, and ensure that (1/QPS) seconds have elapsed since then. For example, for a rate of
   * QPS=5 (5 tokens per second), if we ensure that a request isn't granted earlier than 200ms after

        for (double qps : qpsToTest) {
          // If warmupPermits = maxPermits - thresholdPermits then
          // warmupPeriod = (1 + coldFactor) * warmupPermits * stableInterval / 2
          long warmupMillis = (long) ((1 + coldFactor) * warmupPermits / (2.0 * qps) * 1000.0);
          RateLimiter rateLimiter =
              RateLimiter.create(qps, warmupMillis, MILLISECONDS, coldFactor, stopwatch);

        for (double qps : qpsToTest) {
          // If warmupPermits = maxPermits - thresholdPermits then
          // warmupPeriod = (1 + coldFactor) * warmupPermits * stableInterval / 2
          long warmupMillis = (long) ((1 + coldFactor) * warmupPermits / (2.0 * qps) * 1000.0);
          RateLimiter rateLimiter =
              RateLimiter.create(qps, warmupMillis, MILLISECONDS, coldFactor, stopwatch);

- Kubemark: added two flags, `--kube-api-qps` which indicates the maximum QPS to the apiserver, and `--kube-api-burst` which indicates maximum burst for throttle to the apiserver. ([#124147](https://github.com/kubernetes/kubernetes/pull/124147), [@devincd](https://github.com/devincd))

- Fix throughput when scheduling daemonset pods to reach 300 pods/s, if the configured qps allows it. ([#124753](https://github.com/kubernetes/kubernetes/pull/124753), [@sanposhiho](https://github.com/sanposhiho)) [SIG Scheduling]

  </mime-type>
  <mime-type type="application/vnd.proteus.magazine">
    <glob pattern="*.mgz"/>
  </mime-type>
  <mime-type type="application/vnd.publishare-delta-tree">
    <glob pattern="*.qps"/>
  </mime-type>
  <mime-type type="application/vnd.pvi.ptid1">
    <glob pattern="*.ptid"/>
  </mime-type>
  <mime-type type="application/vnd.pwg-multiplexed"/>