Linux containers run using the same Linux kernel of the host (machine, virtual machine, cloud server, etc). This just means that they are very lightweight (compared to full virtual machines emulating an entire operating system).

This way, containers consume **little resources**, an amount comparable to running the processes directly (a virtual machine would consume much more).

title: TensorFlow, Large-scale machine learning on heterogeneous systems

# etcd V3 Quickstart Guide [![Slack](https://slack.min.io/slack?type=svg)](https://slack.min.io)

etcd is a distributed key value store that provides a reliable way to store data across a cluster of machines.

## Get started

### 1. Prerequisites

- Docker 18.03 or above, refer here for [installation](https://docs.docker.com/install/).

### 2. Start etcd

fi
# VERY basic check to ensure the [and-cuda] package variant is installable.
# Checks TFCI_BAZEL_COMMON_ARGS for "gpu" or "cuda", implying that the test is
# relevant. All of the GPU test machines have CUDA installed via other means,
# so I am not sure how to verify that the dependencies themselves are valid for
# the moment.
if [[ "$TFCI_BAZEL_COMMON_ARGS" =~ gpu|cuda ]]; then

### Create a backup job

#### Backup Virtual Machines with Veeam Backup and Replication

- Under Home > Jobs > Backup in Navigation Pane, click on Backup Job button in the ribbon and choose Virtual Machine. Follow the on screen wizard.

- On the Storage screen, choose the Scale-out Backup Repository that was configured previously.

 *
 * By orchestrating two different servers with the same port but different IP addresses, we can
 * test what OkHttp does when both are reachable, or if only one is reachable.
 *
 * This test only runs on host machines that have both IPv4 and IPv6 addresses for localhost.
 */
@Timeout(30)
class FastFallbackTest {
  @RegisterExtension
  val clientTestRule = OkHttpClientTestRule()

                                    + now
                                    + ", lastUpdated = " + lastUpdated + "). Please verify that the clocks of all"
                                    + " deploying machines are reasonably synchronized.");
                    versioning.setLastUpdated(now);
                    changed = true;
                }
            }
        }

        if (changed) {

    -   Uses `pycpp.sh`, `code_check_changed_files.sh`

These "env" files match up with an environment matrix that roughly covers:

-   Different Python versions
-   Linux, MacOS, and Windows machines (these pool definitions are internal)
-   x86 and arm64
-   CPU-only, or with NVIDIA CUDA support (Linux only), or with TPUs

## How to Test Your Changes to TensorFlow

MinIO in distributed mode lets you pool multiple drives (even on different machines) into a single object storage server. As drives are distributed across several nodes, distributed MinIO can withstand multiple node failures and yet ensure full data protection.

## Why distributed MinIO?

### Server Memory

For example, if your code loads a Machine Learning model with **1 GB in size**, when you run one process with your API, it will consume at least 1 GB of RAM. And if you start **4 processes** (4 workers), each will consume 1 GB of RAM. So in total, your API will consume **4 GB of RAM**.

And if your remote server or virtual machine only has 3 GB of RAM, trying to load more than 4 GB of RAM will cause problems. 🚨