To summarize and conclude this series, I want to offer some “rules” or at least “guidelines” that I believe small and mid-sized businesses should consider when planning for their virtual infrastructure.
First – ensure your design provides data and compute redundancy across multiple “boxes” wherever possible. And Second – you can provide for both with a single “mirror” copy of your data simply by putting your “RAID” mirror on a remote server. These both seem rather obvious but it’s amazing how often these simple principles are ignored … often by putting all the storage inside a single shared SAN “box” that becomes a single point of failure, or creating costly over-replication … using both local RAID with storage overhead to protect against a disk failure and in addition creating additional replicas on different hosts to protect against server failures, often resulting in 4 or more copies of all your data.
Next, be very cautious purchasing “de-featured” or “entry-bundled” software. Many vendors of software only solutions will have multiple “editions” and bundles that may seem attractively priced until you need to expand and add a new server or activate some new feature they figured you ultimately will need. Seemingly “expected” things can instantly double your licensing costs such as adding one additional server to a cluster exceeding a “bundle” or wanting to use one “enterprise feature” (perhaps like storage live migration after you’ve maxed out a non-scale out SAN controller and need to move data between multiple SANs).
This may be controversial but I suggest you save admin training days / $’s for applications that create real business value for your company – not basic infrastructure and IT plumbing. I believe that will benefit you as an IT professional. Along that line, the level of typical training required as well as install / configuration time of a solution is a good representation of complexity and cost of ongoing maintenance you should expect – something that takes 10x longer to setup and configure is going to take at least 10x more time to maintain, patch and troubleshoot. Quite honestly maybe more than 10x because installation is a lot more “standard” than keeping a complex, changing, interdependent environment up and running.
Here is one that few will argue yet many ignore – don’t buy today what you think you will need in 3 years. I understand that many architectures don’t lend themselves to expansion …or perhaps expansion requires using “today’s” CPU model or architecture for compatibility and there are concerns about future availability and costs. The best strategy is to select an architecture that avoids that (hint, hint) and even then buy just enough to cover your ability to predict your needs. Three years is way to much for anyone, 6 months to a year may be reasonable for most and fit normal purchasing / budgeting cycles.
Lastly – simplicity is good. Dealing with fewer vendors that offer standardized modular configurations is way better than assembling the very best but totally customized mouse trap you can. Not only should you have peace of mind knowing that Scale gives you one place to call, but when you do call we know your exact configuration. We test it with every software change we make and provide you updates for the entire software stack in one step.
I hope this series has been beneficial and would love to hear any additional “rules” you would like to suggest in the comments.
Please complete the form below and someone from Scale Computing will be in contact with you within 24hrs.×