Database Encryption, Part 7: Wrapping Up.By Adrian Lane
In our previous posts on database encryption, we presented three use cases as examples of how and why you’d use database encryption. These are not examples you will typically find cited. In fact, in most discussions and posts on database encryption, you will find experts and and analysts claiming this is a “must have” technology, a “regulatory requirement”, and critical to securing “data at rest”. Conceptually this is a great idea, as when we are not using data we would like to keep it secure. In practice, I call this “The Big Lie”: Enterprise databases are not “data at rest”. Rather the opposite is true, and databases contain information that is continuously in use. You don’t invest in a relational database just to have a place to store your data; there are far cheaper and easier ways to do that. You use relational database technology to facilitate transactional consistency, analytics, reports, and operations that continuously alter and reference data.
Did you notice that “to protect data at rest” is not one of our “Three Laws of Data Encryption”?
Through the course of this blog series, we have made a significant departure from the common examples and themes cited for how and why to use database encryption technologies. In trying to sift through the cruft of what is needed and what benefits you can expect, we needed to use different terminology and a different selection process, and reference use cases that more closely mimic customer perceptions. We believe that database encryption offers real value, but only for a select number of narrowly focused business problems. Throwing around overly general terms like “regulatory requirement” and “data security” without context muddies the entire discussion, makes it hard to get a handle on the segment’s real value propositions, and makes it very difficult to differentiate between database encryption and other forms of security. Most of the use cases we hear about are not useful, but rather a waste of time and money.
So what do we recommend you use?
Transparent Database Encryption: The problem of lost and stolen media is not going away any time soon, and as hardware is often recycled and resold – we are even seeing new avenues of data leakage. Transparent database encryption is a simple and effective option for media protection, securing the contents of the database as it moves physically or virtually. It satisfies many regulatory requirements that require encryption – for example most QSA’s find it acceptable for PCI compliance. The use case gets a little more complicated when you consider external OS, file level, and hard drive encryption products – which provide some or all of the same value. These options are perfectly adequate as long as you understand there will be some small differences in capabilities, deployment requirements, and cost. You will want to consider your roadmap for virtualized or cloud environments where underlying security controls provided by the external sources are not guaranteed. You will also need to verify that data remains encrypted when backed up, as some products have access to key and decrypt data prior to or during the archive process. This is important both because the data will need to be re-encrypted, and you lose separation of duties between DBA and IT administrator, two of the inherent advantages of this form of encryption. Regardless, we are advocates of transparent database encryption.
User Level Encryption: We don’t recommend it for most scenarios. Not unless you are designing and building an application from scratch, or using a form of user level encryption that can be implemented transparently. User level encryption generally requires rewriting significant chucks of your application and database logic. Expect to make structural changes to the database schema, rewrite database queries and stored procedures, and rewrite any middleware or application layer code that talks to the database. To retrofit an existing application to get the greater degree of security offered through database encryption is not generally worth the expense. It can provide better separation of duties and possibly multi-factor authentication (depending upon how you implement the code), but they normally do not justify a complex and systemic overhaul of the application and database. Most organizations would be better off allocating that time and money into obfuscation, database activity monitoring, segmentation of DBA responsibilities within the database, and other security measures. If you are building your application and database from scratch, then we recommend building user level encryption in the initial implementation, as this allows you to avoid the complicated and risky rewriting – as a bonus you can quantify and control performance penalties as you build the system.
Tokenization: While this isn’t encryption per se, it’s an interesting strategy that has recently experienced greater adoption in financial transaction environments, especially for PCI compliance. Basically, rather than encrypting sensitive data, you avoid having it in the database in the first place: you replace the credit card or account number with a random token. That token links back to a master database that serves as the direct tie to the transaction processing system. You then lock down and encrypt the master database (if you can), while only using the token throughout the rest of your infrastructure. This is an excellent option for distributed application environments, which are extremely common in financial and retail services. It reduces your overall exposure of by limiting the amount and scope of sensitive data internally, while still supporting a dynamic transaction environment.
As with any security effort, having a clear understanding of the threats you need to address and the goals you need to meet are key to understanding and selecting a database encryption strategy.