Redundant or not Redundant, how hard do you try?
by admin on Jun.27, 2006, under Datacenter
I have a question to pose to everyone and I will give my take. What is redundancy? If you use the term redundant and datacenter in the same sentence how true is the statement. If you have two web servers that are load balanced but both of them point back to one database is that redundant? If you have multiple power supply each plugged into its own UPS but no generator is that redundant? If the core systems like cooling and power aren’t redundant how hard should you work at making the systems redundant? How hard should you work on the front end to make it redundant if the backend is a single point of failure? My option is that none of those situations are redundant, maybe load sharing but not redundant. If something as simple as A/C and power is not redundant in a datacenter I wouldn’t worry about spending thousands of dollars and add a ton of complexity for something that will break anyway. I would like comments on this, am I way off base? The question I pose to people with a “redundant” system without backup power or A/C is what is more likely? A Cisco switch/router burning out or the power going out? What about the A/C unit freezing up and causing the datacenter to overload, that sure happens a lot.
July 5th, 2006 on 9:10 pm
I think that you must look at what failure points are serviceable in the time allowed and take the appropriate steps to assure that you achieve the best results with the budget that you have. CRAC Units when designed properly should be redundant in at least two units for the datacentrer – I feel that many of today’s issues revolve more or less on the demands that are placed on the environments presented within the last two – three years. Massive arrays of disk teamed with smaller form factors and blades all contribute to the increased power consumption thus leading to cooling issues. When stress is placed on any one point and not built and designed for the additional load its only a matter of time. 1 -1.5KW was not uncommon per rack only three years back, now companies like HP and APC are expecting cooling requirements to pass the 80KW per rack area with cooling pushed to the blade itself and moving away from forced air.
UPS Systems and CRAC units can be redundant, but even the most redundant power plan will require separate power legs into the building at opposite ends taking diverse paths back to the head end not to mention the load and all circuits must be transferable. Added is a great paper on this design and the relevancy of the five tier approach to datacenters published by the Up Time Institute http://www.upsite.com/tuipages/whitepapers/tuitiers.html
IT is a mix of life experiences, problem solving, budgeting and overall willingness to think on your toes.
What is the likely hood of a UPS unit going down compared to a server? Who knows, it depends what power grid you are on compared to how critical your server is.
I think that you are on the right track – I think that a great topic you should talk about would be the transition of business needs from tier definitions. The theoretical design of a fault tolerant datacenter is a promising thing, but the overall issue is balancing such a want with the underlining business need at the right time for the business.
As some of your readers published already, not many companies have a large bank roll for hardware. With that being said a judgment call is made. I have a feeling (most of the time) redundancy is built to offset human error – that Cisco power supply may not go down, however someone may unplug by mistake.
My 2 cents….