I’ve mentioned that the SNIA SDC Conference provided the catalyst for writing this blog series on data protection. (See part 1 and part 2.) While there was a lot of discussion at the Conference on protecting data from loss, there were also discussions about encryption of data at rest. I noticed that the conversations about data protection and encryption were separate and distinct. Networks also employ data encryption methods to protect data in transit. I was sitting in one of the sessions when I began to wonder how many times a data block is encrypted and decrypted in its life. I also wondered how many times encrypted data is encrypted again as “data at rest” becomes “data in transit.”
Data at rest encryption can be divided into two categories: media encryption and object encryption. Media encryption focuses on protecting data at the physical device level such as the disk drive or tape. Encryption for data at rest can be performed at multiple places in the data storage stack—host, SAN appliance or target device (tape drive, tape library, disk drive). There are use cases for each deployment option, but generally it’s best to encrypt as close to the physical media as the use case allows. There are often trade-offs to be examined. For instance, encryption defeats the value of deduplication. In most data sets, there is a lot of repeated data that can be managed more efficiently if deduplicated. If host-based encryption is employed, the value of deduplicating data downstream, such as WAN acceleration, is eliminated.
The benefit of encryption at the media level has generated interest in encrypting tape drives and Self-Encrypting hard Drives (SED). Tape use cases are pretty straight forward: create a data copy on media that can be shipped off-site for protection of the data at the primary site. Tape used to be the primary backup media, but with long recovery times and the data explosion, tape has been relegated to tertiary copies and long term archive of data that doesn’t need to be online. The key is tape is designed to be shipped off-site, meaning the shipment could be hijacked. Encrypting the data on the tapes makes a lot of sense. A box of encrypted tapes has the same value as a box of used tapes, i.e., not worth hijacking.
I have “sold” a lot of SEDs in my career. I’ve always tried to be honest with customers. SEDs have limited value in data center operations. Drives deployed in the data center aren’t intended to transport data out of the data center. (There are a couple of valid use cases where moving data on drives makes sense, such as data center moves or seeding a disaster recovery site, but of all the drives deployed, very few are used in this manner). I’d often test a customer’s view of SEDs with a simple question, “Do you have locks on your data center doors?” Some customers would get the joke and I knew I could have a frank conversation. If the customer didn’t get the joke (i.e., understand that SEDs only provide protection if the physical drive falls into the wrong hands), I proceeded cautiously. Two other factors come into play at that moment: there was an account manager in the room who is commission driven (SEDs are slightly more expensive than non-SEDs) and paranoid customers with deep pockets are an IT supplier’s best friend. Of course, just because you’re paranoid, doesn’t mean that there aren’t a few thousand hackers out there looking to gain value from your data.
Bottom line is, if someone gains access to your network or manages to compromise someone’s username/password, SEDs don’t help. The encryption key is automatically applied to the drive when the drive is started. Any process that has access to the system after that isn’t going to be denied access by the drive encryption.
The primary value of SEDs is when a drive is deployed outside the data center, where the primary or secondary data protection is at the drive. Best example is a laptop. You can assume that the data on a stolen laptop without encryption is in the hands of the thief. Other portable devices, such as tablets, smartphones, etc., also have encrypted storage devices, though not a hard drive in a conventional sense. Note: Many Solid State Drives (SSDs) are also SEDs, which makes the case for the SSD option in your next laptop stronger.
Before my friends in the storage industry start tweeting about me, I do see a few values for SEDs in the data center.
- Compliance. Many security offices require SEDs—it never hurts to have SEDs, just understand where they fit in the security stack.
- Storage is going to be repurposed. Lab environments, cloud providers, etc., where storage may be used for one project or customer today and another tomorrow may desire or require compete data erasure. The easiest way to erase data is to encrypt it and then delete the key. The data will technically be on the drive, but not accessible.
- End of life destruction. Drives (spinning and SSD) do wear out and need to be disposed of. Some people require physical destruction of the drive (heard of people shooting them with a high-powered rifle, but never witnessed it). There are services that will crush or shred the drive. However it’s easier to shred the key. (Paranoid people do both.)
Object level encryption is another way to address protecting data at rest. I’m using a very vague definition of object for this discussion. Objects are often associated with a particular cataloging mechanism that supports very large numbers of items to be stored. In this case, I’m not being that specific. Think of an object as something that could be addressed at the application level. I spent a great portion of my storage career working on Networked Attached Storage (NAS) systems, another poorly named technology. NAS is essentially a purpose-built file server. For this conversation a file and the file system could be considered objects.
I’ve had many conversations with customers about protecting files with encryption. Customers often wanted to encrypt the entire file system. This is pretty straight forward: one key to manage for all of the content in the file system. The problem is the one key tends to be widely distributed—any person or process that needs access to a file gets the key to the entire file system. A side effect of this kind of a solution is that all metadata of the file system is also encrypted. So operations like backups that operate based on the creation and modification timestamp need to have the key to the file system. Therefore systems like backup servers, anti-virus servers, etc., have to be secure as they literally have the keys to the kingdom.
Another approach is to encrypt the files within the file system. Think of systematic zipping of files with encryption and a password. This has the benefit of not affecting the metadata. A file can be moved, emailed, deleted, etc., without decrypting the file. The backup software doesn’t need to have the key to execute, and the files in the backup are encrypted. Operations that need to access the internals of the file, such as anti-virus or full-text search still require the keys. The challenge is managing the keys and access control lists. Some files are written and read by only one person/application. However, most files are intended to be shared. For instance, emailing an encrypted file doesn’t do the recipient any good unless you also provide the key. I know a lot of people who encrypt any file that they put in their “free cloud storage.” It isn’t that they don’t trust the cloud provider—it’s just that sometimes a little paranoia is a good thing.
So why not encrypt everything everywhere? As I pointed out above, encrypted file systems are hard to manage. Encryption also makes it harder to detect intrusions in the network when the data-in-transit is encrypted. I can remember pointed discussions between the storage admins and the network admins about encrypting replicated data. The storage admin wanted the data encrypted at the source array and decrypted at the target array. The network admin wanted to encrypt at the WAN edge device, so they had visibility into the data leaving the building.
An interesting shift is the use of encryption by hackers. Rather than copy your data, they encrypt it and then offer to sell you the key. This phenomenon is called ransomware. While detection of the malicious code is the preferred defense, a good data backup enables a good backup plan. Suppose you have hourly copies of your data. Rather than choose to pay the ransom, you could choose to restore your data to the point in time before your data became encrypted.
At this point, if you’re expecting me to tie a nice little bow around data protection, you’re going to be disappointed. Protecting data in a world where the threats were application errors, failed components, undetected bit swaps and natural disasters was a challenge. Today, the threats are using teams of well-funded experts focused on finding the weak links in your data security structure. The threat landscape is constantly changing. It is very difficult, if not impossible, to protect against all threats. The IT technology industry is working to provide a solution component. However, the threat volatility forces overall protection to be reactive to the threat technology.
Organizations need to look at the problem the way the Navy does when protecting data.
- Implement layers of security
- Assume that any layer is penetrable
- Minimize the damage of a penetration
First step, limit access to your data infrastructure through identity checks and limit access to need to know. Avaya Identity Engines provide a powerful portfolio of tools for managing user and device access to your network. However, assume that someone will figure out how to forge credentials and gain access to your infrastructure.
Avaya SDN Fx provides a key foundational component of a data security solution, minimizing the exposure of your network to unauthorized access. So when the spy gains access to your network, you can limit the exposure and keep the perpetrator from wandering around your network looking for the good stuff.
Data in transit and data at rest encryption and data backups provide another level of defense and recoverability when other layers are breached.
Finally, everybody needs to be involved in keeping data secure. I was interrupted while writing this conclusion to help a sales engineer with an opportunity. I emailed him several docs and links to others as background information. Even though the docs are marked as to which ones were for internal use only, I noted in the email which docs were sensitive and couldn’t be shared with the customer. Proper strategies include systems, processes, and people all working together across organizations and technology stacks to prevent data from being lost or ripped off.
I’ve always been a believer that the effort to make things idiot proof was often wasted because they just keep making better idiots. In this case, they’re making better experts to come after your data. Fortunately, we have very intelligent experts working for the good guys too. We’ll always be one step behind, but we can continue to strive to minimize the threat surface and minimize the impact of the surface being violated.