Why Global Dedupe Is The ‘Killer Feature’ of Cloud Backup

Why Global Dedupe Is The ‘Killer Feature’ of Cloud Backup

Thanks to recent advances in mobile technology, device storage capacity continues to closely follow Moore’s Lawthat is, doubling approximately every two years. Magnetic disks in laptop devices first reached terabyte (TB) capacities a few years ago, and now flash-based disks have broken the TB barrier. With the rise of the mobile workforce, much of today’s work is increasingly being performed on mobile devices, and protecting this data is just as critical as if it were stored in the data center.

For the past few years, Gartner has been tracking advances in end user data protection in its periodic report on Critical Capabilities for Enterprise Endpoint Backup. During that time, content created on mobile devices has become increasingly media-rich and hence, documents, presentations, and even emails have grown in size. The increase in data on devices, coupled with unpredictable network connections, means the ability to protect data effectively is a function in how timely it can be collected for recoverability. When it comes to data protection of mobile devices, this means the scale of data that needs to be backed up every cycle rises exponentially.

Over the past decades, improvements in hard drive storage have roughly paralleled increases in bandwidth, according to Nielsen’s Law of Internet Bandwidth, at roughly a 50% increase per year. As content creation moves to mobile devices such as smartphones and tablets, content types are evolving beyond text-based emails and documents to rich media content like video, powerpoint or pdfs. In tandem, the volume of unstructured data within an enterprise is exploding, according to IDC. Clearly, the epicenter of business is moving to distributed devices, workers and networks.


By virtue of their inherent mobility, mobile devices are more likely than attached storage devices to backup their data over a network; and today’s devices have multiple options: office Wi-Fi, a hotspot at the local coffee store, even a telco data plan. Although backup software can be configured to use a specific method, waiting for your chosen network to become available exposes you to significant risk. Imagine if your VP were to lose his or her laptop in an airport terminal, moments after putting the final touches to your latest sales presentation! It should be obvious that when devising a backup strategy, you need to weigh the cost of Wi-Fi access or data plan overages against the very real possibility of losing your critical data forever.

One of the keys to ensure the data is protected is to use an efficient algorithm that not only compresses but also removes data redundancies so the minimal amount of data is transferred for the maximum amount of data protection. To accomplish this, vendors utilize deduplication, but not all deduplication is the same. Understanding their differences can aid in ensuring your organization meets its data protection objectives.

So, what are the various types of deduplication?

Deduplication at Target: Inefficient for Network but Efficient on Storage Savings

With deduplication at target, you copy your data to a storage device, the storage processor identifies the duplicate data, and maintains only a single copy—all other copies are discarded. Now, whenever you need to access the relevant data you get it from this single copy.

Deduplication at target can be performed in either inline or offline mode. In the case of inline deduplication, as new data is being received by the storage processor, a ‘fingerprint’ of the data is compared with existing fingerprints in real time. If the same data already exists on the target, it is skipped and only a reference to the existing data is maintained. In offline mode, data is first copied to a target; an offline scan then removes duplicated data and maintains a single copy. All references to the data then point to this single copy.

Consider a situation where you have multiple copies of a file, each slightly different from one another (e.g., a work that is in progress). Using deduplication at target, each device will copy its version of the file over a network layer to the backup store. Whether in online or offline mode, the storage processor in the backup store will then scan these files, find the unique content in each and delete all but one of these unique parts, maintaining a single copy of all the data. However, this creates significant overhead as each device’s version of the file must initially be sent over the network. It is highly inefficient in terms of bandwidth but conducive to overall storage savings, as only a single copy per target is maintained.

What if you could identify whether the data has already been stored at the target before copying, and skip sending it altogether with just a pointer to the stored copy? This is called deduplication at source.

Local Deduplication at Source: Saving Network Bandwidth at the Cost of Endpoint Resources

In the case of local deduplication at source, data is scanned locally first, followed by the identification of any unique data, which is then backed up once. Subsequent backups need only include a reference to the original data, thus conserving bandwidth. While there is the advantage to only sending unique data across the network once, scanning for and comparing this information locally can be resource-intensive and weigh down heavily on the CPU and memory of a mobile device. For very large datasets, this can be highly inefficient for an endpoint device.

Further, this only ensures that deduplication happens across just a single device at a time.

Now, this is alright for your personal laptop, but for business organizations with multiple copies of data spread across multiple devices, duplicate data across these devices gets transferred multiple times.

Global Deduplication at Source: The Best of Both Worlds

The answer to the dilemma is global deduplication at source. Using this method, the data ‘fingerprint’ is calculated at the source; this fingerprint is then sent across to the target where it is compared to existing data. If a match exists, regardless of its source, only a reference to the data will be copied. By ‘global, we mean across all users and their devices. The two earlier methods are either user or device only and therefore never get the network-effect that global delvers, which significantly lowers the overhead of data being transferred.

Take this scenario for example: a file is sent as an email attachment to a group of coworkers. With the deduplication at target method, each mobile device will copy the file over a network layer to the target, where all but one of these files is deleted and a single copy maintained. Using the local deduplication at source method, once a file is sent to the backup store, no other copies will be sent by that device and only the metadata of the file with a unique identifier will be transferred, but multiple devices will send that file each time as it’s unque within that device. In the case of Global deduplication at source, a file is sent across network only once, be it for 10 devices or 100s.

As the number of users deployed increases, the duplicate data begins to accumulate. For successive users, only a fraction of their entire backup set data will need to be uploaded. This reduces bandwidth usage and improves the speed of deployment as it is rolled out across the organization.

In the real-world example below, a very large deployment in a global consulting company was implemented within a single business quarter thanks to global deduplication at source. While the users had nearly 300 TB of total backup data, less than 150 TB of data was ultimately transferred to the backup cloud. This was made possible by deduplicating not only on a per-user basis, but globally across the entire organization. This 50% data transfer savings applies to the initial deployment, and it increases to 80% over time as you only backup unique data.


In this real-world customer example, of 300 TB of total backup data, less than 150 TB was ultimately transferred due to global data deduplication.

The global deduplication at source method, however, scales beyond a single user to the organizational level. The result is that you may not even need to copy that file over a network to the backup store, as someone else may have already sent it there as a part of their backup cycle. This can save a significant amount of network traffic at the cost of finding a unique identifier for that file.

The decision often boils down to what is cheaper: CPU cycles or network bandwidth. However, across an entire organization we have seen this method save an incredible amount of bandwidth over time—in one real-world case, nearly 15TB over the course of one year! Without the benefit of global deduplication at source, nearly twice the amount of data would need to be transferred to backup enterprise data. Not only does this significantly increase bandwidth consumption, it also increases the duration of backups cycles and reduces RPO in the event of a disaster recovery situation.


The above example from a technology company illustrates that for a fully deployed customer the deduplication advantage persists even when data transferred is growing slower than data backed up growth.

In the case of locally deduplicated backups, you will see an incremental increase in bandwidth, as each device will have a fixed amount of duplicate content. With global deduplication, however, the bandwidth savings accelerate as more devices are included. Initial backups also become quicker, as each subsequent user enjoys the deduplication benefits of previous device backups.

Comparison of Deduplication Methods for Endpoint Backup


At Target Deduplication

At Source Local Deduplication

At Source Global Deduplication


Endpoint Device resources

Low resource use required

High resources used. All of the dedup computation is done locally. Involves fingerprint calculation of data and comparison.

Moderate resources used for fingerprint calculation for the data.

Bandwidth Savings

No B/W savings at all. All the data is transferred to backup store

Moderate bandwidth savings. For data globally duplicated such as mail attachments etc. the same data is sent multiple times over network

High bandwidth savings as only single copy of data is sent over network.

Storage footprint at the backup store

Low storage footprint as only single copy is maintained

Moderate storage savings as each device backs up deduplicated data only.

High storage savings as globally single copy of data is stored

Backup server resources

High resources use required for deduplication processing

low resource use required as deduplication happens on endpoint device.

Moderate resource use required for inline deduplication processing.


Finally, by observing a few simple best practices, global deduplication at source can offer even greater advantages to an organization. For example:

  • Geographically-dispersed organizations can speed up backup of their remote users by first deploying their head office users who have access to better bandwidth, and then later deploying the remote users who may benefit from deduplication with existing data.
  • Mobile or tethered devices can be deployed later in the implementation cycle, potentially saving bandwidth on those devices, which typically backup either on Wi-Fi or other wireless networks.

To learn more about data deduplication, check out our previous blog posts here and here.

Get a free trial of Druva’s single dashboard for backup, availability, and governance, or check out these useful resources:

View Gartner’s 2015 Endpoint Backup Critical Capabilities Report to see how the players stack up.


* Critical Capabilities for Enterprise Endpoint Backup, Pushan Rinnen Et. al, 24 October 2013 and 09 October 2012
Gartner does not endorse any vendor, product or service depicted in its research publications, and does not advise technology users to select only those vendors with the highest ratings or other designation. Gartner research publications consist of the opinions of Gartner’s research organization and should not be construed as statements of fact. Gartner disclaims all warranties, expressed or implied, with respect to this research, including any warranties of merchantability or fitness for a particular purpose.

Ashish Karnik

I started Druva's Performance Engineering team about 2 years ago. Previous to Druva I was working with Symantec (formerly and currently VERITAS). My primary focus area is storage and distributed systems.


Leave a reply

Your email address will not be published. Required fields are marked *