bias problem. And because it makes pretty flame graphs!

Let user processes read performance stuff:
```
sudo bash
echo 0 > /proc/sys/kernel/kptr_restrict
echo 1 > /proc/sys/kernel/perf_event_paranoid
exit
```

Grab the async profiler from https://github.com/jvm-profiling-tools/async-profiler
and run `prof async` like so:
```

  TF_DeleteStatus(s);
  return NULL;
}

// A compute function. This will never actually get called in this test, it's
// just nice to know that it compiles.
void compute(void* kernel, TF_OpKernelContext* ctx) {
  TF_Tensor* input;
  TF_Status* s = TF_NewStatus();
  TF_GetInput(ctx, 0, &input, s);
  TF_DeleteTensor(input);
  TF_DeleteStatus(s);
}

// Exercises tensorflow's C API.

   *       then sequentially accessing it could result in other processes dying. This is solvable
   *       via madvise(2), but that obviously doesn't exist in java.
   *   <li>Ordinary copy. Kernel copies bytes into a kernel buffer, from a kernel buffer into a
   *       userspace buffer (byte[] or ByteBuffer), then copies them from that buffer into the
   *       destination channel.
   * </ol>
   *

# MinIO Server Limits Per Tenant

For optimal production setup MinIO recommends Linux kernel version 4.x and later.

## Erasure Code (Multiple Drives / Servers)

| Item                                                            | Specification |
|:----------------------------------------------------------------|:--------------|
| Maximum number of servers per cluster                           | no-limit      |

issues with Docker 1.9.1 can be found below.
  * CPU hardcapping will be enabled by default for containers with CPU limit set,
if supported by the kernel. You should either adjust your CPU limit, or set CPU
request only, if you want to avoid hardcapping. If the kernel does not support
CPU Quota, NodeStatus will contain a warning indicating that CPU Limits cannot
be enforced.

// and return "non-ready" handles in `retvals`. Note that any handles contained
// in 'op' should not be mutated till the kernel execution actually finishes.
//
// For sync execution, if any of the inputs to `op` are not ready, this call
// will block till they become ready and then return when the kernel execution
// is done.
// TODO(agarwal): change num_retvals to int from int*.

		_ = setMaxResources(globalServerCtxt)
	})

	// Verify kernel release and version.
	if oldLinux() {
		warnings = append(warnings, color.YellowBold("Detected Linux kernel version older than 4.0 release, there are some known potential performance problems with this kernel version. MinIO recommends a minimum of 4.x linux kernel version for best performance"))
	}

	maxProcs := runtime.GOMAXPROCS(0)

  // TODO(tfrt-devs): Figure out a way to deprecate following features after
  // migrated to TFRT.
  //===--------------------------------------------------------------------===//
  // Clear pending nodes in thread executors and kernel caches.
  virtual void ClearCachesAndThreadExecutors() = 0;

  // Initialize the step resource container for a training step. This is used
  // in current TF runtime. For tfrt, it is used by fallback op handler.

func fixLongPath(path string) string {
	// Do nothing (and don't allocate) if the path is "short".
	// Empirically (at least on the Windows Server 2013 builder),
	// the kernel is arbitrarily okay with < 248 bytes. That
	// matches what the docs above say:
	// "When using an API to create a directory, the specified
	// path cannot be so long that you cannot append an 8.3 file

	if err != nil {
		buf, err = os.ReadFile(cgroupV1MemLimitFile)
	}
	if err != nil {
		return 0
	}
	limit, err = strconv.ParseUint(strings.TrimSpace(string(buf)), 10, 64)
	if err != nil {
		// The kernel can return valid but non integer values
		// but still, no need to interpret more
		return 0
	}
	if limit >= 100*humanize.TiByte {
		// No limit set, or unreasonably high. Ignore
		return 0
	}
	return limit
}