Over the past couple of months I have run into quite a few customers looking for “true” highly available architectures, and explaining to me that their small business needs this feature.
HA can and does mean different things to different companies, as it should. When looking at what type of security your specific company needs there are a few key pieces to look at:
- How much data can you afford to lose? Realistically, if you lost 5/10/15 minutes worth of input could your company survive? How much data loss would be catastrophic?
- How much does your downtime cost you? $10/100/1000 per hour? For every minute you are down how much are you costing the business?
- What is the max amount of time you can be down? At what point will you be unable to resume business if you are down for x days/weeks?
To the questions above: the less data loss you can absorb and the less time you can be down the higher the cost to the business. So, the best strategy is to understand the costs associated with each and find where the ROI makes the most sense. By understanding the needs of the business, and not those of the IT department, you can start planning the correct HA architecture.
Site level HA for the SMB
Clustering – I have seen clustering done in multiple ways. The most common is through server level clustering in which you duplicate writes to two servers and synch them together with a heartbeat. In the case one of the servers (primary) goes down the second server can take over. This can work well for very small customers without a ton of processing or storage needs, but as the business grows this architecture does start to get expensive as you have to purchase twice the number of servers/processors/software licenses as is needed for your business. Additionally, it means you will not be utilizing ½ of the processing power that was purchased by IT.
3-2-1 – This is the most common HA for the SMB market today and relies on the idea of having 3+ servers (with virtualization), 2 SAN switches, and 1 SAN. In this way the data resides on a HA SAN (99.999% uptime) and the applications reside on the servers. If one server goes down, the applications are automatically migrated from the bad server to one that is up, spreading the processing across two servers instead of three. This architecture can be less expensive as customers are able to consolidate the number of servers they are managing, but there is some downtime as applications are migrated (a few seconds).
A trend I have seen lately is to combine the two theories on HA above into a single solution that involves clustering two SAN’s together at a single location. The thought process is to protect against a SAN level failure. This idea is being pushed largely by small SAN manufactures that do not have the same type of HA built into their systems and so cannot provide their customers the correct level of protection. Most IT environments today can provide businesses with 99.99% uptime. Today’s enterprise storage systems typically have between 99.99% and 99.999% uptime, meaning they have more HA built into their single architectures than IT can build in the rest of their IT. Enterprise storage arrays were built specifically to withstand any singular part breaking. This is one of the major benefits for customer’s looking to consolidate their infrastructure. By investing in a technology that is in itself highly available, businesses can then focus on other IT initiatives than spending money duplicating a Storage system.