Wednesday, November 7, 2012

Disaster Recovery Fundamentals


Now that we have defined DR in our last post, let's look at some of the options for creating a DR plan for your business or your customer.

In a disaster situation, both the users AND the data need to be shifted from one computing environment to another. This effectively means there are two DR plans that need to be enacted simultaneously. If you are only providing colocation to the client, then your only role is really to keep the facility from having an outage in the first place. Even with that limited role, it is still very important to speak to your customers about their own DR plans. At a minimum, it shows the customer you care about their business. And perhaps, you can also provide them white label or partner services that are of value to you and your company, as well as your partner ecosystem. Where things get complicated is if your company provides *some* level of managed services, but may not be responsible for the entire computing environment, including network or application. In such scenarios, your company is now part of the customer Operations team and will need to know exactly which services to provide in a DR situation and in which order, as well the triggers that put them in motion and their own feedback loop to the client when their portion of the work is complete. Yes, you are in the middle of a potential panicked situation. The last scenario, and least likely, is that your company provide full Operations support for the customer and owns the entire computing environment, as well as DR plan. This is least likely as the customer almost always owns at least the application and potentially the users, if they are corporate users and not Internet users.

The first half of our solution involves shifting users from one computing environment to another. Shifting Internet users is trivial now thanks to geoloadbalancing technologies that have matured in the last 15 years. In simple terms, this means there are two potential computing environments for the Internet user. The geoloadbalancing system will determine if one environment is available and the other is not available. It will then send Internet users to the available environment. Of course, there are a lot of bells and whistles that will allow tuning based on geography, network latency, computing environment load, and many other factors if both environments are available, but again, that is more of a Business Continuity scenario. For DR, let's just think of it as a light switch where an environment is ON or OFF, with only one environment being ON at a time and Internet users directed to that ON environment. Managed Services and network companies should be selling this capability. The second user scenario involves corporate or VPN users. Remote corporate users many times use VPN clients as a way to connect to a computing environment, usually for access to corporate file servers or email. The good news is that a VPN client can easily accommodate connections to differing computing environments, many times with the ability to automatically try secondary connections if the primary connection fails. This is called "rollever" and just means if computing environment A is not reachable, the VPN client will automatically try to connect to computing environment B. From an infrastructure standpoint, this simply means both computing environment A and computing environment B must have the ability to allow the corporate user VPN connection and to authenticate it. Simple tasks for your IT department. The scenario of an office connecting to the computing environment is very much the same as the VPN user. The only difference is that instead of a VPN client for each user, the office will have one large VPN tunnel that serves all users. In the case computing environment A is not avialable, the router or other VPN device will automatically "rollover" to computing environment B. In reality, the VPN device is continuously trying to connect to both environments and will just use the one that is available.

And now the real work - shifting data from one environment to another. It would be so nice if real time synchronous replication worked perfectly and was cheap, but this is just not the case. The engineers can explain the problems with trying to synchronize data between multiple environments, but from a business perspective, this just means about twice the cost of operation to keep mirrored environments available - twice the hardware, twice the space, twice the power, and maybe if you are lucky, you only have to pay network at one location. Synchronous replication is also more of a Business Continuity plan, so again, let's keep this as DR and the light switch - ON and OFF. In reality, if computing environment B is a physical coloction, the customer will always have the servers on and ready and maybe just not have data replicated. In such a scenario, the DR strategy for the data will be to move it from one location to another. This is usually done with block level replication, meaning chunks of data are transferred. The problem comes with just how much data is to be transferred and how fast it needs to be there. Remember, if computing environment A is down completely, a customer may not have any access to that data. Then how do we do it? By continuously replicating the data to another location. This is known as Storage-As-A-Service or Backup-As-A-Service when using a 3rd party. This is just replication if a client is using their own infrastructure. The data continuously gets backed up, then in a DR situation, is transferred again to computing environment B from the data backup location. Again, this has to be architected based on the amount of data and the time a customer must have it replicated to have computing environment B available. This is your SLA.

Virtualization to the rescue. There are a number of cloud service providers that will allow virtual machines to remain dormant (OFF) and provide a guarantee of how long it will take to turn on each virtual machine. This would be computing environment B. A customer may be able to only pay for storage without colocation or machines and then pay for computing cycles when the virtual machines are powered ON. Think of it - 1,000 virtual machines turning on in less than a minute. We still have the issue of getting the data to the virtual machine, but the customer now can save a massive amount of money for that secondary computing environment. Does your company offer this service? If not, you really should be looking to partner as a cohesive DR plan presented to a client is going to be of more value.

While these are just the basics of DR, hopefully it will provide you with more of an understanding and way to speak with your customer, as well as ideas for partnering in your ecosystem.

No comments:

Post a Comment