- Improved scheduling latency when many gated pods ([#124848](https://github.com/kubernetes/kubernetes/pull/124848), [@gabesaba](https://github.com/gabesaba)) [SIG Scheduling and Testing]

			runtime.Gosched()
			acc = e.cached.Load()
		}
	}
	atomic.AddInt64(&acc.N, 1)
	atomic.AddInt64(&acc.Total, int64(value))
	atomic.AddInt64(&acc.Size, sz)
}

// total returns the total call count and latency for the last minute.
func (e *lockedLastMinuteLatency) total() AccElem {
	e.mu.Lock()
	defer e.mu.Unlock()
	return e.lastMinuteLatency.getTotal()
}

- `kubelet` now emits a metric for end-to-end pod startup latency, including image pull. ([#121041](https://github.com/kubernetes/kubernetes/pull/121041), [@ruiwen-zhao](https://github.com/ruiwen-zhao))
- `kubelet` now exposes latency metrics of different stages of the node startup. ([#118568](https://github.com/kubernetes/kubernetes/pull/118568), [@qiutongs](https://github.com/qiutongs))

- Added a warning log, an event for cgroup v1 usage and a metric for cgroup version. ([#125328](https://github.com/kubernetes/kubernetes/pull/125328), [@harche](https://github.com/harche))
- Added apiserver.latency.k8s.io/apf-queue-wait annotation to the audit log
  to record the time spent waiting in APF queue. ([#123919](https://github.com/kubernetes/kubernetes/pull/123919), [@hakuna-matatah](https://github.com/hakuna-matatah))

	smallFileThreshold = 128 * humanize.KiByte // Optimized for NVMe/SSDs

	// For hardrives it is possible to set this to a lower value to avoid any
	// spike in latency. But currently we are simply keeping it optimal for SSDs.

	// bigFileThreshold is the point where we add readahead to put operations.
	bigFileThreshold = 128 * humanize.MiByte

- Improved scheduling latency when many gated pods ([#124851](https://github.com/kubernetes/kubernetes/pull/124851), [@gabesaba](https://github.com/gabesaba)) [SIG Scheduling and Testing]

    checkpoints continue to be readable; controlled by the `write_version`
    argument, `tf.train.Saver` now by default writes out in the new V2 format.
    It significantly reduces the peak memory required and latency incurred
    during restore.
*   Added a new library for library of matrix-free (iterative) solvers for
    linear equations, linear least-squares, eigenvalues and singular values in