Transfer Appliance is a rugged storage box that Google sends to your office or data center, so you can move huge amounts of data to Google Cloud by post and not through the internet.

Most data reaches the cloud through the internet. For a few photos or a small database, that is perfect. But when a company wants to move hundreds of terabytes, or even a petabyte, the internet becomes the slowest part of the plan. A terabyte is about a thousand gigabytes, roughly the size of two hundred thousand typical photos. A petabyte is a thousand terabytes.

Transfer Appliance solves this by turning the problem upside down. Instead of pushing data through a thin network pipe, you rent a physical device from Google. It arrives at your site, you copy your files onto it over your fast local network, and you ship it back. Google then loads the data into your Cloud Storage bucket from inside its own high-speed network. Ordering the device, copying the data, sending it back and finding your files in the cloud is the whole story.

What it is good at

• Moving very large data sets when the network is too slow, too busy or too costly to use.
• Starting a cloud project, such as a data lake, with years of old data already in place.
• Moving data out of sites with poor internet, such as remote research stations or factories.
• Giving a security-minded company a controlled, encrypted, trackable way to move data.

What it is not

Transfer Appliance is not a database migration tool, a backup program or a live sync service. It copies files and objects in bulk, once, to Cloud Storage. Ongoing changes after the copy must be moved by other tools. Later chapters explain how teams combine the box with online transfer tools.

Data grows faster than networks. The arithmetic of moving it is simple, and often shocking.

Moving data takes time equal to the amount of data divided by the speed of the connection. If the numbers are small, this is invisible. When the numbers are large, it dominates the whole project. The table below shows the best case, where the connection runs at full speed without any pause. Real networks are slower, because they are shared and have overhead.

Mbps means megabits per second, and Gbps means gigabits per second. A company with a one gigabit link that is shared by hundreds of employees may only be able to spare a small part of it for a transfer. Using the whole link for weeks may also slow down email, video calls and customer systems.

The hidden problems of a long online transfer

A short list of words is enough to understand the whole subject. Learn them here once.

Cloud Storage, bucket and object

Cloud Storage is Google Cloud’s service for storing files at huge scale. It keeps each file as an object, which is the file’s contents together with its name and some information about it. Objects live inside buckets, which are named containers you create. A bucket is similar to a top-level folder, although Cloud Storage does not have real folders inside it. The folder-like paths you see are only part of object names.

Project and organization

A project is the container in which your Google Cloud resources, permissions and bills live. Your Transfer Appliance order belongs to a project, and so does the bucket where the data will land.

Appliance

The appliance is the physical box. It holds a large amount of storage, has network ports, and acts like a big network drive on your local network while it is at your site.

NFS

NFS, short for Network File System, is a common way for computers to share folders over a network. Transfer Appliance offers its storage as an NFS share, so your servers can mount it and copy files to it with ordinary tools, much like using an external drive.

Finalization

Finalization is the step in which you tell the appliance that copying is finished. It then closes and locks the data, so nothing can be added or changed, and it gets ready for shipping.

Rehydration

Rehydration is Google’s name for the process that reads your data from the returned appliance and writes it into your Cloud Storage bucket. The name comes from the idea that the data, packed and encrypted on the box, is brought back to life in the cloud.

Checksum

A checksum is a short code calculated from the contents of a file. If even one bit of the file changes, the checksum changes too. Systems compare checksums before and after a copy to prove that nothing was damaged on the way.

Encryption and keys

Encryption scrambles data so that it is unreadable without a secret key. Transfer Appliance encrypts your data on the box, and you control the key that unlocks it.

Transfer Appliance is one of several ways to get data into Google Cloud. Choosing well saves weeks and money.

The online options

Storage Transfer Service is a managed service that copies data into Cloud Storage from other clouds, from web addresses, from other buckets and, with a small agent program, from file systems in your own data center. It runs in the cloud, handles retries, and can be scheduled. Command-line tools such as the gcloud storage commands are good for smaller jobs that you start yourself. Dedicated network links, such as Cloud Interconnect or a VPN, give you a faster or safer path for online transfers.

The offline option

Transfer Appliance moves data physically. Its strength is in volume, and its price is time spent waiting for shipping, plus some manual work at your site.

flowchart TD
  S["Need to move data to Google Cloud"] --> Q1{"Would the online transfer take about a week or more?"}
  Q1 -->|No| ON["Use online tools such as Storage Transfer Service or gcloud storage"]
  Q1 -->|Yes| Q2{"Can you add bandwidth or a dedicated link affordably?"}
  Q2 -->|Yes| ON2["Consider Interconnect or more bandwidth, then transfer online"]
  Q2 -->|No| Q3{"Is an appliance available for your location?"}
  Q3 -->|Yes| TA["Use Transfer Appliance"]
  Q3 -->|No| ALT["Plan a staged online transfer or other options"]

The system has a few simple parts. Seeing how they connect makes every later step easy to follow.

flowchart LR
  SRC["Your data on file servers and storage arrays"] --> NET["Your local high speed network"]
  NET --> TA["Transfer Appliance at your site"]
  TA --> SHIP["Shipping carrier"]
  SHIP --> UP["Google upload facility"]
  UP --> GCS["Cloud Storage bucket in your project"]
  GCS --> USE["BigQuery and other Google Cloud services"]
  CON["Google Cloud console order and tracking"] -.-> TA
  CON -.-> UP

Part 1: your source data and servers

This is the data you want to move, stored on file servers, network storage devices or ordinary computers. The machines that read this data are usually the true limit on copy speed, so they deserve attention.

Part 2: your local network

The appliance connects to your network with high-speed ports. Because the appliance is next to your servers, the transfer runs over your fast internal network and does not touch the internet. This is why a copy that would take months online can take days.

Part 3: the appliance

The box contains storage, a computer that serves the NFS share, encryption features and monitoring. It records what you copy and prepares the data for the return trip.

Part 4: the carrier and the upload facility

A shipping company carries the appliance both ways. At Google, trained staff receive it, connect it to a fast network and begin rehydration. The upload goes directly into the bucket that you named in your order.

Part 5: the console

You place the order, follow its status, and receive notifications in the Google Cloud console. It is also where you find instructions and, in some cases, tools for your copy.

Choosing the right size of appliance is the first real decision. Too small, and you must order again. Too large, and you pay for space you do not use.

Google has offered Transfer Appliance in more than one size. Models have included smaller devices holding tens of terabytes, and larger ones holding hundreds of terabytes, with names such as TA40 and TA300 in more recent generations. Older generations used different names and capacities. Because the lineup has changed over time, check the current documentation for the models, the usable capacity, the physical size and the network speed of each.

Raw versus usable capacity

Raw capacity is the total size of the disks inside. Usable capacity is the space you can really fill with your files, which is smaller because some space is used for protection against disk failures and for the system. Always plan with usable capacity.

Do not count on compression

Some storage products advertise bigger numbers through compression and deduplication, which shrink data by removing repeats. Data such as videos, photos and already-compressed archives hardly shrinks at all. Plan as if your files will take their full size, and treat any saving as a bonus.

How to size your order

The cheapest problems are the ones you find before the box arrives. A short checklist avoids most surprises.

Data readiness

• Know what you are moving. Make a list of folders, sizes, file counts and owners. Decide what will not be moved.
• Check file names and sizes. Cloud Storage has limits on object names and sizes. Find problem files early. The chapter on files, names, sizes and metadata explains the main limits.
• Plan for changing data. Decide whether files will keep changing during the transfer, and how you will catch up afterwards.
• Clean first. Delete junk, temporary files and duplicates you no longer need.

Cloud readiness

• A project with billing. The order lives inside a Google Cloud project, and billing must be active.
• A destination bucket. Create the bucket, choose its location and storage class, and confirm the permissions needed for the upload service.
• The right people and permissions. Someone with the proper role must place the order, and others need view access to track it.
• A key plan. Decide how encryption keys will be created, stored and backed up, before the box arrives.

Site readiness

• Space and power. The appliance needs rack space or a safe table, suitable power and cooling. Check the technical sheet for exact needs.
• Network ports and cables. Make sure you have switch ports of the right speed, and cables or adapters that fit.
• Receiving the box. Decide who will accept the delivery, where it will be stored, and how access is controlled.
• Servers that can feed the box. Check that your source machines can read data quickly and have the NFS client software installed.

People readiness

Name a project lead, a person for the network, a person for the storage and servers, and a person who understands the cloud side. Agree who will watch the copy, who will verify the results, and who decides when source data may be deleted.

From the first click to the last email, a Transfer Appliance job follows a clear series of stages.

Cloud permissions decide who can order the box, who can watch it, and which service may write into your bucket.

Google Cloud uses IAM, short for Identity and Access Management, to give people and programs permission to act. For Transfer Appliance, a few kinds of permission matter. Exact role names and steps can change, so use the console instructions as your final guide.

People who order and manage

A person who places and manages orders needs a Transfer Appliance administrator role in the project. Other team members can be given viewer roles, so they can follow progress without being able to change or cancel the order. Grant these roles to groups where possible, so that access can be reviewed easily.

The destination bucket

The upload service must be allowed to write objects into your bucket. The console usually guides you in granting the right permission to a Google-managed service identity. Grant only what is needed, on that one bucket, and not across the whole project.

Bucket choices

• Location. Choose a region or multi-region close to your future users and compliant with your data rules. Changing the location later means copying again.
• Storage class. Choose Standard for data in frequent use, and colder classes for data that will rarely be read. The chapter on storage classes and bucket design explains the choice.
• Access control. Prefer uniform bucket-level access, so permissions are simple and consistent.
• Protection. Consider object versioning or retention settings if the data must not be deleted or overwritten for some time.

The box will sit in your own building for days or weeks. A little attention to space, power, cooling and network avoids frustrating delays.

Physical space

Depending on the model, the appliance may be designed to fit in a standard server rack, or may need a stable surface in a secure room. Check the dimensions and weight in the documentation, and make sure the delivery route, doors and lifts can handle the shipping case. Keep the original packaging, since you will use it to ship the device back.

Power and cooling

The device draws electrical power and gives off heat. Use the type of power connection that is specified, and place it in a room with enough ventilation. Do not block the air vents, and avoid damp or dusty places.

Network

Connect the appliance to a switch that runs at the speed the appliance supports. Using slower ports wastes the box’s capability. Decide how the box will get its network address, either automatically or with a fixed address that you assign. Make sure firewalls between your servers and the box allow the NFS traffic and any management access that you need.

Physical security

The appliance holds your data. Keep it in a locked room with controlled access and, if your policy requires it, camera coverage. Keep a log of who enters and who handles the box, from arrival to departure.

The copy happens on your own network, so a little network knowledge turns a slow week into a fast weekend.

Speeds in plain words

Network speeds are measured in bits per second. A gigabit link carries one billion bits each second, which is roughly 125 megabytes per second at best, once you convert bits to bytes. A 10 gigabit link carries ten times more. Many offices still use one gigabit connections on desk machines and servers. If your servers connect at one gigabit, then even a very fast appliance will wait for them.

These are ideal numbers. In real life, expect lower speeds because of protocol overhead, small files, shared switches and slow disks.

Where the bottleneck hides

• Server network cards. A server with a one gigabit card cannot send faster than that, no matter how big the appliance link is. Use faster cards, or use many servers.
• Switches and uplinks. A fast switch connected to the rest of the network by a slow cable becomes a narrow road for all traffic.
• Cables and optics. Damaged or wrong cables cause errors and retransmissions, which slow everything.
• Disks and storage arrays. The source storage must read fast enough. Older arrays under heavy business load may be the true limit.
• The file system and mount settings. Poorly tuned mount options and tiny transfer sizes can waste a fast link.

Measure before you start

Before the big copy, run a test that moves a large file from a source server to a test target on the same switch, and measure the rate. Then run the same test from several servers at the same time. This tells you the total capacity of your path, and it shows whether adding more servers will really help.

The day the box arrives is the day the clock starts. A calm, ordered setup avoids early mistakes.

Copy speed depends less on the appliance and more on how well your servers feed it. Parallel work is the secret.

The appliance exposes your storage as an NFS share. Your servers mount the share and copy data into it with the usual tools for copying files, such as the standard copy and synchronization utilities of your operating system. For huge data sets, one copy command running on one server will not fill a fast connection. You need many copies running at the same time.

The three limits to watch

• The network. A 10 gigabit link can carry roughly one gigabyte per second at best. Slower ports, shared links and poor cables lower this.
• The source storage. Old disks or busy servers may read slowly. If your source system is the limit, adding appliances will not help, but spreading reads across more servers may.
• The file pattern. Millions of tiny files copy much more slowly than a few large files, because each file has overhead. Handling many small files in parallel reduces this cost.

flowchart TD
  A["Copy is slow"] --> B{"Is the network link the limit?"}
  B -->|Yes| B1["Use faster ports and good cables, avoid shared links"]
  B -->|No| C{"Is the source storage slow?"}
  C -->|Yes| C1["Spread reads over more servers or disks"]
  C -->|No| D{"Too many tiny files or too few parallel streams?"}
  D -->|Yes| D1["Run several copy streams in parallel and group the work by folder"]
  D -->|No| E["Check CPU, mount settings and the appliance health indicators"]

Good copying habits

• Split the work by folder. Give each server or copy process its own set of folders, so streams do not collide.
• Use several servers. Many modest servers copying in parallel often beat one powerful server.
• Keep a log. Save the output of each copy job, so errors can be traced and re-run.
• Do not change files in place. If a file changes during the copy, you may capture a half-old, half-new file. Copy from a snapshot or a quiet moment where possible.
• Track progress. Compare the amount copied with the total, and estimate the finish time, so you can warn people about delays.

A file system and Cloud Storage are different worlds. Knowing the differences prevents surprises after the upload.

Names

Cloud Storage stores each file as an object whose name can be up to 1,024 bytes long. File systems often allow different lengths and different characters. Names with unusual characters, such as control characters or certain symbols, can cause trouble. Very long folder paths can also exceed the limit once the path becomes part of the object name. Scan your source for such names before you copy.

Sizes

A single Cloud Storage object can be as large as 5 tebibytes. Files bigger than this cannot be stored as one object, and must be split before transfer. Most organizations have few such files, but video archives and database dumps sometimes do.

Folders

Cloud Storage has a flat structure. The slashes in object names create the illusion of folders. Empty folders usually do not carry over as objects, so if an empty folder matters to your applications, plan to recreate it or note it separately.

Permissions and metadata

File systems keep owners, groups, permissions and special attributes, which are not stored in the same way in Cloud Storage. Access in Cloud Storage is controlled through IAM and bucket settings. If your applications depend on file permissions, plan how to express the same rules in the cloud. Also, special items such as symbolic links, which are pointers to other files, may not be preserved as such. Check the current documentation for what is kept, and what is not.

Timestamps

Modification times may be kept as metadata in some workflows and not in others. If a timestamp matters to your business, test it with a sample and confirm that it arrives as you expect.

Not all data is a folder of ordinary files. Some kinds need special care before they go in the box.

Live databases

Copying the files of a running database is dangerous. The files change constantly, and a copy taken at the wrong moment may be unusable. Instead, use the database’s own tools to create a consistent backup or export, and copy that file. Remember that the appliance puts files in Cloud Storage, which is not a place to run a live database. The data must later be loaded into a database service.

Virtual machine disks

Large disk images of virtual machines can be copied as files, but they must be consistent. Shut down the machine or use a snapshot, and then copy the image. Images may be larger than the limit of a single object, in which case they must be split. After the upload, special import tools can turn images into machines in the cloud.

Backup archives

Backup software often writes data in its own special format. Moving those files to the cloud keeps them safe, but reading them later may need the same backup software, or a cloud-based version of it. Check that your backup product can restore from a copy stored in Cloud Storage before you rely on it.

Tape libraries

Data on tape is not on a file system. To use an appliance, the tapes must first be read and restored to disk, which can take a long time. Count that time in the plan. Large tape archives are a common reason for starting early and working in waves.

Many small files

Millions of tiny files copy slowly because of the overhead per file. If you control the data, consider packing small files into larger archives before copying, then unpacking in the cloud when needed. If you must keep each file separate, use many parallel streams.

When data travels on a physical box, you must be able to prove that what arrives is exactly what left.

Data can be damaged in many small ways: a bad cable, a failing disk, a copy that stops early, or a file that changed during the copy. Most of the time nothing goes wrong, but when it does, the damage may stay hidden for months. The cure is to check the data with checksums at several stages.

Where checks happen

What to compare after the upload

• The number of files. The count in the bucket should match the count at the source, apart from files you intentionally skipped.
• The total size. A difference may point to a missing folder or a failed upload.
• Checksums for key data. Cloud Storage stores checksums for each object. Compare them with checksums calculated at the source for important files.
• The error and skip logs. Read every line. Files that failed must be re-sent by another method.

Your data should be unreadable to anyone who finds or steals the box. Encryption makes that true, and you hold the key.

Transfer Appliance encrypts the data that you copy onto it. Without the correct key, the contents look like random noise. The key is not stored on the box in a form that anyone can use, and you control it, either by providing a passphrase or by using a key held in a key management service. Options and steps can change, so follow the current instructions.

flowchart TD
  F["Files copied to the appliance"] --> E["Encrypted on the appliance with your key"]
  E --> L["Locked after finalization"]
  L --> R["Shipped back sealed"]
  R --> D["Decrypted only during upload, using your key access"]
  D --> B["Stored in your Cloud Storage bucket"]

Why key care matters

Because you hold the key, you also hold the risk. If the key is lost, nobody can decrypt the data on the appliance, and the transfer cannot be completed. If the key is leaked, someone who also gets the box could read it. Treat the key as one of the most valuable items in the project.

Key handling rules

• Create and store it securely. Use your normal secret storage or key management system, with access limited to a few people.
• Keep a backup. Store a second copy in a different protected place, such as a safe, with a documented procedure for retrieving it.
• Separate the key from the box. Never ship the key with the appliance, and never write it on the packaging.
• Record who has access. Keep a list, and review it when the project ends.
• Plan the end of the key’s life. After rehydration is complete and verified, follow your policy for retiring the key used for the transfer.

The return trip needs as much care as the arrival. A clear process protects both the data and the schedule.

Finalization

When all data is copied and your copy logs are clean, you finalize the appliance. Finalization closes the data so that nothing new can be added, flushes any data still in memory to the disks, creates the record of what is stored, and prepares the box for transport. Once finalized, you cannot add files. If you find a missing folder afterwards, you must send it by another method, or ask support whether the appliance can be reopened.

Preparing for return

• Power the appliance down properly, and disconnect cables in the order that the instructions give.
• Pack it in its original case or packaging, using the supplied seals.
• Attach the return shipping label, and record the tracking number.
• Take photographs of the packed box and of the seal numbers.
• Hand it to the carrier, or schedule a pickup, and keep a receipt.

Chain of custody

Chain of custody means a written record of every person who has handled the box and every place where it has been, from the moment it left Google until it comes back. It helps if the box is lost, delayed or damaged, and it supports audits for regulated data. Keep a simple log with names, dates, times and actions.

Once the box reaches Google, a careful, mostly automatic process turns its sealed contents into objects in your bucket.

Step by step

What can slow or stop rehydration

• Missing or wrongly set key access, so the data cannot be unlocked.
• Wrong permissions on the destination bucket, so writes are refused.
• Object names or sizes that Cloud Storage cannot accept, which appear as individual failures.
• Quotas or organization policies in your project that conflict with the upload.
• Problems with the appliance itself, which Google handles with the support process.

Seeing all the actors in one picture makes the full journey easy to remember.

sequenceDiagram
  participant You as Your team
  participant G as Google Cloud
  participant C as Carrier
  You->>G: Request an appliance in the console
  G->>C: Ship the appliance
  C->>You: Deliver to your site
  You->>You: Connect the box and copy data
  You->>You: Finalize and lock the appliance
  You->>C: Hand over for return shipping
  C->>G: Deliver to the upload facility
  G->>G: Verify and upload data to Cloud Storage
  G->>You: Notify that data is available
  G->>G: Securely erase the appliance

Where time is spent

Different stages take different amounts of time. Shipping may take days in each direction, depending on distance and customs. The copy at your site may take days or weeks, depending on how well you prepare. Rehydration often takes days, because it moves large volumes, though it runs on fast links. Waiting and approvals add further time.

Where your data lands, and in which class, decides what it costs for years. Decide before the box is sent, not after.

Storage classes

Cloud Storage offers classes that trade price against how often you read the data. Standard suits data used often, with the highest storage price and no extra charge for reading. Nearline suits data read about once a month. Coldline suits data read about once a quarter. Archive suits data read less than once a year, with the lowest storage price but charges for reading and minimum storage periods. Prices and minimum durations change, so check the current pricing page.

Location

A bucket can live in one region, in a pair of regions, or across a large multi-region area. Single regions are cheaper and close to compute in that region. Multi-regions offer more resilience. Choose according to where your users and applications are and any legal rules about where data may be stored.

Lifecycle rules

Lifecycle rules move objects to colder classes or delete them automatically as they age. A common idea is to land data in a class that fits its first months, and let rules move it to colder storage later. Test rules on a small set of objects first.

Bucket layout

• Use meaningful, stable prefixes in object names, so that data is easy to find and secure by area.
• Avoid putting unrelated data in one huge bucket if different teams need different permissions.
• Consider separate buckets for raw imported data and for cleaned data that others will use.

Once your data is in a bucket, a set of features helps you protect it, organize it and keep its cost under control.

Object versioning

With versioning turned on, Cloud Storage keeps older versions of an object when it is overwritten or deleted. This protects against mistakes, at the cost of extra storage. Combine it with lifecycle rules that delete old versions after a time.

Retention policies and holds

A retention policy says that objects cannot be deleted or replaced for a set period. A hold pauses deletion for specific objects, for example during a legal investigation. These tools support rules that require records to be kept.

Lifecycle management

Lifecycle rules move objects to colder classes or delete them based on age and other conditions. They save money without anyone needing to remember to do it.

Autoclass

Autoclass is a feature that moves objects between storage classes automatically according to how often they are accessed. It is helpful when access patterns are hard to predict. Check the current documentation to see how it works and how it is priced.

Access control and sharing

Use uniform bucket-level access and IAM roles to give people and programs the access they need. For short-term sharing, signed links can give temporary access to a single object without creating accounts.

Inventory and insights

Reports that list the objects in a bucket, with sizes, classes and checksums, are very useful after a large transfer. They allow you to compare the cloud copy with your source lists, and to see how your storage grows over time.

When the email says “complete,” the work is not finished. It is the start of a careful final check.

A box full of valuable data traveling by road deserves a clear security plan with several layers.

Layer 1: encryption you control

Data on the appliance is encrypted, and you control the key. Even if the box is lost or stolen in transit, its contents should remain unreadable. This is the most important layer, and it relies on good key handling.

Layer 2: physical protection

The appliance is built to be rugged and to show signs of tampering. At your site, keep it in a locked room with restricted access and a visitor log. During shipping, use the supplied seals and tracked service, and check the seals on arrival and on return.

Layer 3: narrow permissions

Limit who may order, who may see the order, and what the upload service may do in your bucket. Review the roles at the start and end of the project, and remove them when the work is done.

Layer 4: network safety

Place the appliance on a network segment that only the source servers and administrators can reach. Do not connect the appliance to the open internet unless the instructions require it. Use firewall rules to allow only the traffic that is needed.

Layer 5: audit and records

Keep the chain of custody, the copy logs and the audit logs of order actions in the cloud. These records support investigations and compliance checks, and they help you prove that you handled the data properly.

Layer 6: secure erasure

After the upload, Google securely erases the appliance before it is reused. You can ask for confirmation of the process, and your own compliance team may want it in writing.

A good project keeps eyes on three things: where the box is, how the copy is going, and what the cloud side reports.

Order status in the console

The Google Cloud console shows the state of each order, such as requested, approved, shipped, delivered, returned, being uploaded and complete. Notifications are sent to the contacts you named when you placed the order. Make sure those contacts are real, monitored mailboxes, preferably shared team addresses and not a single person’s inbox that may be unattended during holidays.

Copy progress at your site

Track how much data is copied each day. A simple table with the planned and actual volume gives early warning when the copy is slower than expected. Watch the health indicators of the appliance, such as disk status, temperature and network errors, using the management screen or the monitoring features that the instructions describe.

Audit logs in the cloud

Actions such as creating, changing or canceling orders are recorded in Cloud Audit Logs, which answer the question “who did what and when.” Review them if something unexpected happens, and give your security team access to them.

Alerts worth setting up

• A reminder when the appliance has been at your site for most of the free holding period, so you do not pay late fees unknowingly.
• An alert if the copy rate drops below the plan for more than a day.
• A notice when the order status has not changed for longer than expected, so someone can ask support.
• A reminder to revoke temporary permissions and retire keys after completion.

A few simple numbers tell you whether the transfer was a good project, and what to do better next time.

When the work is done, it is tempting to celebrate and move on. A short review with real numbers is worth the effort. It proves the value of the project to the people who paid for it, and it gives your team a head start on the next one.

Measures worth recording

Questions for the review meeting

• Did the actual timeline match the plan, and where did we lose or gain time?
• Which preparation step saved the most trouble?
• Which problem should we have found before the box arrived?
• Was the appliance the right choice compared with the online option?
• What will we change in the checklist for the next project?

Honest numbers early make the difference between a pleasant project and an angry finance meeting.

What you pay for

Pricing details change, so read the current pricing page. In general, the cost has several parts. There is a fee for each appliance order, which covers the device and its handling. There may be a daily charge for the time the appliance stays at your site beyond an included period. Shipping may be included or charged separately depending on the region. Once the data lands in Cloud Storage, you pay normal storage charges for the chosen class. Uploading data into Cloud Storage does not carry the usual charge for data leaving the cloud, because the data is entering, though your own electricity, staff time and network costs are separate.

Comparing against the online alternative

An honest comparison includes more than the appliance fee. For the online route, count extra bandwidth that you must buy, the staff time to babysit a long transfer, the cost of disturbing business traffic, and the cost of waiting. For the appliance route, count the order fee, late fees if any, staff time for the copy, shipping time and the risk of delays.

Controlling the cost

• Return the appliance within the included period, by preparing the copy well in advance.
• Order the right size, since extra devices cost more, but too few devices cause a second round.
• Clean data before the copy so you transfer and store less.
• Choose the right storage class and lifecycle rules, to avoid paying for expensive storage you do not need.

Very large projects use many appliances, sometimes at several sites. Organizing them is a project in itself.

Parallel appliances

If your data exceeds the capacity of a single device, or you want to finish sooner, you can order several appliances and fill them at the same time. This works well when different servers hold different parts of the data. Divide the data into logical parts, such as by department, by date or by folder, and assign each part to a device.

Waves

Another approach is to work in waves: order one appliance, complete the whole cycle, learn from it, and then order the next ones. This is slower but safer for the first time. A pilot wave reveals problems with names, permissions, speed and your own process, before you commit to a large order.

Multiple sites

If data is spread across several offices or data centers, each site needs its own appliance, local preparation and trained staff. Keep a central coordinator who tracks every device, key and schedule, and who makes sure that naming, bucket layout and verification rules are the same everywhere.

The best projects use both worlds. The box moves the mountain, and online tools move the small changes since the copy began.

Large data sets rarely stand still. While your box travels for weeks, new files appear and old files change at the source. If you simply switch over after the box arrives, you will lose those changes. The usual solution is a two-step approach.

Storage Transfer Service is well suited to the delta stage, because it can compare and copy only differences, and it can be scheduled to repeat. Its agent program can read from file systems in your own data center. The small final copy happens in hours, not weeks.

Every route has a price. Look at both sides of Transfer Appliance honestly.

Compared with other approaches

Trade-offs to think about

Time versus effort. The box shortens the data transfer but adds logistics work. If the team has no one to handle the physical part, the plan may stall.

Control versus convenience. Holding the key gives you control and responsibility. If your team cannot manage keys well, build that skill before the box arrives.

One big move versus steady flow. The appliance is for the big move. If your need is a steady flow, a dedicated link and online tools may be a better long-term investment.

Patterns are lessons that other teams paid for. Anti-patterns are the traps they fell into.

Six helpful patterns

Three costly anti-patterns

The habits below are simple. Most of them come from teams who learned the hard way.

Best practices

• Do the arithmetic honestly. Use the real, shared bandwidth you can use, and compare with the full appliance timeline.
• Clean before copying. Delete what you do not need, and avoid moving junk to the cloud.
• Test with a sample. Include odd names, big files and tiny files, and examine how they arrive.
• Prepare the cloud side early. Create the bucket, grant narrow permissions, and decide the key plan before ordering.
• Use many copy streams. Spread the work across servers and folders, and monitor the throughput.
• Keep records. Save the chain of custody, copy logs, manifests and upload reports.
• Keep contacts alive. Use shared mailboxes for notifications, and make sure someone is available to answer quickly.
• Verify, then delete. Never remove the source before sign-off.
• Close the project. Remove temporary permissions, retire the key and write a short lessons-learned note.

Common mistakes and quick fixes

Things sometimes go wrong. A prepared team recovers quickly, and an unprepared team loses weeks.

Principles for recovery

• Keep the source until the end. Every recovery path relies on the original data still existing.
• Contact support early. A short message on day one saves long waits later.
• Do not improvise with the hardware. Follow the official steps for opening, resetting or repacking the device.
• Write everything down. Times, names and messages help every later conversation.
• Learn and adapt. After any incident, update your checklist for the next appliance.

When data crosses a street, nobody minds. When it crosses a border, laws and contracts may have something to say.

Many organizations are bound by rules about where data may be stored, who may handle it, and how its movement must be recorded. These rules may come from laws, from industry standards or from customer contracts. Moving data physically makes some of these rules more visible, because the box itself crosses places on a map.

Questions to ask before ordering

• Where may this data live? Choose a bucket location that satisfies your rules, and check that the upload facility and route are acceptable.
• May the box cross borders? If your site is in one country and the upload facility is in another, shipping may involve customs, export rules and extra paperwork. Check with your legal team and Google about the route that applies to your location.
• Is the data classified? Highly sensitive data may need approval, extra encryption, special handling or a decision that the appliance route is not suitable.
• What evidence will auditors want? Prepare chain-of-custody logs, key management records, access lists and the upload report in a form that can be shown later.
• Does personal data need special care? Review privacy rules before copying records that identify people, and remove data that you have no right or need to keep.

Documents to keep

Different organizations use Transfer Appliance for different reasons. Find the story that looks like yours.

The wider picture

Offline data transfer is an old idea in a new form. Before the internet, people mailed tapes. Cloud providers now offer managed, encrypted and tracked versions of the same approach, and other major cloud providers offer similar devices for their own platforms. The lesson is simple: bandwidth is not growing as fast as data, so physical movement remains a smart tool for the very largest jobs. We describe the examples above as general patterns, and we do not claim that any named company uses the service in a specific way.

Let us follow a made-up infrastructure engineer named Leena as she moves a large archive to Google Cloud.

Leena works for a company that publishes educational videos. Over twelve years, the company has collected about 300 terabytes of raw and finished footage on network storage in its office. The company wants to move this archive to Google Cloud, so that new editing and search tools can use it. The office has a one gigabit internet link, shared by two hundred people.

Leena does the arithmetic. At full speed the link could move 300 terabytes in about 28 days, but the office can spare only a quarter of the link. That means more than three months, with constant risk of failures and complaints about slow video calls. She decides to look at Transfer Appliance.

First she cleans up. With the editors, she removes duplicate exports and abandoned projects, reducing the plan to 240 terabytes. She scans file names and finds a few thousand files with odd characters and some very long folder paths. She fixes them on the source. She creates a project, a bucket in a nearby region with the Archive class for the oldest footage and Standard for recent work, and narrow permissions for herself and two colleagues.

She orders appliances after reading the current model information, and chooses a quantity that holds the data with a safety margin. She prepares a quiet room with a lock, switch ports at the right speed, and a plan to run copy jobs from six file servers at once. The security team creates the key and a backup in their secret system, and rehearses the recovery.

The box arrives. Leena photographs the packaging, logs the serial number, and connects it. A test copy of a mixed sample works well. She then starts parallel copy jobs, each assigned to a different set of folders, and watches the daily progress. After eleven days, the copy completes. She reads the logs, finalizes the appliance, packs it with the supplied seals, and hands it to the carrier.

Google receives the box and begins rehydration. Leena grants the key access when asked. The upload finishes a few days later, with a small list of failed files caused by names she missed. She re-sends them with an online tool, and runs a delta sync for the footage added in the past weeks. She compares counts, sizes and checksums by folder, and the data owners sign off. Only then does the company plan the retirement of the old storage, and she removes the temporary permissions and retires the key.

A transfer project touches storage, network, security, cloud and legal teams. Clear roles prevent dropped balls.

Roles

Name a project lead who owns the schedule and decisions. Name a storage owner who understands the source data and runs the copy. Name a network owner who prepares ports and monitors throughput. Name a security owner who manages keys, access and physical protection. Name a cloud owner who prepares the project, bucket and permissions. Name a data owner from the business who decides what moves and signs off the result.

Rhythm

Hold a short planning meeting before ordering, a daily short check-in during the copy, and a review meeting after verification. Keep a shared document with the schedule, risks and decisions. A single place for information prevents confusion when many people are involved.

Decision points

• Go or no-go for ordering, after the readiness checklist.
• Go or no-go for finalization, after the copy logs are clean.
• Go or no-go for deleting the source, after verification and sign-off.
• Close-out of permissions, keys and the project, after final acceptance.

You do not need to master everything at once. Move from simple and safe to detailed and powerful.

Keep a notebook of every question that came up and how you answered it. After a first project, that notebook becomes your own handbook, and the second project will feel far calmer than the first.

Short, honest answers to the questions beginners ask most often.

If you remember only a few ideas from this long guide, make them these.