One major concern for many companies is the security of a system. In this context security is defined as protection of data, and quality of service (QOS). Both of these fall under security because a system must protect itself from these two forms of attack. When a company loses a system, it loses business, and the money it loses can reach scary amounts.
For the purpose of this discussion, we will address the first aspect (protection from crackers and system infiltration). Many times, the same fixes for this type of protection inherently fix the other (QOS). System security has three distinct concerns that have been well defined: authentication, authorization, and architecture. Why didn't I say encryption? It is because encryption does not equal security. It is only part of the whole system.
Authentication is the process of identifying parties. In a completely trusted system, authentication is as simple as grabbing a reference to the object or entity you want to manipulate. However, sometimes you have trusted systems running in hostile environments (e.g. the Internet). In those cases you have to make sure that the entity is the correct one.
There are two sides of authentication. On one side, you are authenticating an entity you want to do business with. On the other side, there is an entity wanting to do business with you. This is best understood when you bring it into a financial realm. Let's say you want to get a loan from XYZ bank. If you initiate the exchange at their facility, you have already authenticated them--although you will have to jump through hoops to authenticate yourself. If a telemarketer calls from ACME Loan Machine, Inc., they know who you are--although if you are smart, you will make them jump through hoops to authenticate themselves.
Currently, Avalon identifies blocks as trusted parties. In other words, the fact that they are installed correctly in the proper directory is enough to mark them as authenticated. Much like a good friend--you know them by their voice. Because of the architecture of Avalon, blocks define their own authentication system. That means that whether the block assumes all components are trusted or it forces the blocks to authenticate themselves is up to the block.
For systems that deal with sensitive information, authentication becomes more of an issue. You must validate that the components in use are the ones you have authorized for use. The more open a system is, the more authentication and authorization play an important role. If you have components that are transmitted over some form of communications system (wire, air, etc.) you must verify that they are who they say they are.
For this purpose, protocols like Kerberos and Leighton-Micali have been defined. It is beyond the scope of this page to discuss the pros and cons of each system.
Once a component has been authenticated (in other words its identity has been authoritatively validated), then you must decide what access it has to your systems. A fully trusted component can do what it likes. An untrusted component must be denied functions it is not allowed.
Authorization is part of the architecture, but is significant enough to warrant its own concern. The best rule of thumb is to hide access to everything unless the component is trusted. When hiding is not an option, then you have to resort to denying access. For the security conscious, all denied accesses should be logged along with the component's identifier.
Avalon already has the framework for the proper secure architecture. With proper planning, security can be added without recoding the components. However there are some rules of thumb for paranoid programming.
Throughout my short career I have been exposed to a wide range of security implementations. Having worked for a DoD contractor, I learned about physical security requirements (how to manage paper files, and the rules and regulations regarding secure information). However in a recent attempt to install a solution at our customer's site, I learned more about computer security than I have at any other time.
Our customer, a well known corporation whom I will call ACME for anonymity, is in custody of sensitive information. Product movement data, financial data, and demographic data are all stored in one of a couple of huge server farms. Our application was to reside in the De-Militarized Zone (DMZ), and talk to a database behind the firewall. When they discovered that the application server we chose required Remote Procedure Calls (the subject of many CERT advisories in the Unix world), we hit a brick wall. Our application was no longer trusted, and was not to be installed. Luckily the next version of the application server fixed that problem.
In the midst of this, I decided to learn as much as I could about operating in a hardened environment. Hardening involves turning off access to anything that is not needed, and only allowing what is needed. All mail servers, ftp servers, rpc services, telnet, and any other clear text communications ports need to be removed. All communication into and out of the box needs to be encrypted, with the exception of HTTP (although we were using HTTPS).
All of this was necessary. I understand the principles behind it. However, I decided to ask their security analyst how to learn more--so we don't run into this problem again. He gave me a link to SANS institute and a word of advice: "Be paranoid. Don't trust anything."
There is a point where the aforementioned philosophy goes a bit too far. The case in point relates to a story I heard working with the DoD. During the height of the "Cold War" between the US and Russia, a US commanding officer required that all the computers be placed on these huge power conditioners to reduce the Electro-Magnetic Interference (EMI) coming from the power line feedback of the computers. His rationalization was that he was positive KGB agents had the technology that they could read the EMI interference on the power lines and read the bits on the computers. In hindsight this is improbable; no one can identify the bits travelling on a computer from EMI in the power lines due to the level of noise (clocks, addressing lines, encrypted data, and other computers on the network).
The security industry has a general rule of thumb to ensure that you don't spend too much time in this area: The cost of cracking the server should be more than the information that is protected is worth. Basically it boils down to this: sites like Slashdot and SourceForge do not require as much security as a bank. Protecting free information is not as critical as protecting people's bank accounts. I am not minimizing the contributions of the aforementioned sites to the world as a whole, but people become much more agitated if their bank account is wiped clean than if their news article is defaced. This is not to say that if you are protecting free information then you can safely ignore security. While the information is monetarily free to customers, it has value that can be measured in pride, respect, and accountability.
The moral of the story is be paranoid to a point. Avalon is designed to be a trusted system, and will be improved to work in an untrusted network.