measure: rx.measure,
		N:       rx.N + o.N,
	}
}

func calcAvg(x, y float64, n1, n2 int64) float64 {
	if n1+n2 == 0 {
		return 0
	}
	avg := (x*float64(n1) + y*float64(n2)) / float64(n1+n2)
	return avg
}

// Add a new transfer
func (rx *XferStats) addSize(sz int64, t time.Duration) {
	if rx.measure == nil {
		rx.measure = newRateMeasurement(time.Now())
	}
	rx.measure.incrementBytes(uint64(sz))

  /**
   * A map of contents pre-created before experiment starts to only measure map creation cost. The
   * implementation for the creation of contents is independent and could be different from that of
   * the map under test.
   */
  Map<Element, Element> contents;

  /** Map pre-created before experiment starts to only measure iteration cost during experiment. */
  Map<Element, Element> map;

  private void weaklyReferenceQueue() {
    frq = new FinalizableReferenceQueue();
    queueReference = new WeakReference<>(frq.queue);

    /*
     * Queue and clear a reference for good measure. We test later on that
     * the finalizer thread stopped, but we should test that it actually
     * started first.
     */
    reference =
        new FinalizableWeakReference<Object>(new Object(), frq) {

      label: "Orthogonality: How does this change interact or overlap with existing features?"
      description: "Is the goal of this change a performance improvement? If so, what quantifiable improvement should we expect? How would we measure it?"
    validations:
      required: false

  - type: textarea
    id: learning-curve
    attributes:
      label: "Would this change make Go easier or harder to learn, and why?"

  - type: textarea

			return
		}
	}
}

func (r *ReplicationStats) updateMovingAvg() {
	r.RLock()
	for _, s := range r.Cache {
		for _, st := range s.Stats {
			st.XferRateLrg.measure.updateExponentialMovingAverage(time.Now())
			st.XferRateSml.measure.updateExponentialMovingAverage(time.Now())
		}
	}
	r.RUnlock()
}

// ActiveWorkers returns worker stats
func (r *ReplicationStats) ActiveWorkers() ActiveWorkerStat {

   * @param offset the offset into the array at which to start
   * @return the value found in the array in the form of a long
   */
  static int load32(byte[] source, int offset) {
    // TODO(user): Measure the benefit of delegating this to LittleEndianBytes also.
    return (source[offset] & 0xFF)
        | ((source[offset + 1] & 0xFF) << 8)
        | ((source[offset + 2] & 0xFF) << 16)

          if ((hash1 & (1 << k)) == (hash2 & (1 << k))) {
            same[k] += 1;
          } else {
            diff[k] += 1;
          }
        }
      }
      // measure probability and assert it's within margin of error
      for (int j = 0; j < hashBits; j++) {
        double prob = (double) diff[j] / (double) (diff[j] + same[j]);
        assertThat(prob).isWithin(epsilon).of(0.50d);

    RateLimiter rateLimiter = new SmoothWarmingUp(stopwatch, warmupPeriod, unit, coldFactor);
    rateLimiter.setRate(permitsPerSecond);
    return rateLimiter;
  }

  /**
   * The underlying timer; used both to measure elapsed time and sleep as necessary. A separate
   * object to facilitate testing.
   */
  private final SleepingStopwatch stopwatch;

    RateLimiter rateLimiter = new SmoothWarmingUp(stopwatch, warmupPeriod, unit, coldFactor);
    rateLimiter.setRate(permitsPerSecond);
    return rateLimiter;
  }

  /**
   * The underlying timer; used both to measure elapsed time and sleep as necessary. A separate
   * object to facilitate testing.
   */
  private final SleepingStopwatch stopwatch;

You could put an **arbitrary number** to target, for example, something **between 50% to 90%** of resource utilization. The point is that those are probably the main things you will want to measure and use to tweak your deployments.