When purchasing a server, most companies select a server class computer from a reputable manufacturer. And in this day, usually the servers come loaded with redundant components to optimize server availability and make it more resilient. And yet a lot of these servers fail at the first glitch simply because they are not configured properly. Here is a brief blueprint on how to optimally utilize the purchased and paid redundancy.
First, let's analyze what is usually redundant in a server. If we take into account only the garden variety commercial servers and ignore the hugely expensive fault tolerant machines, here is what you usually get:
- Redundant Disk drives
- Redundant Power Supplies
- Redundant Network Adapters
To achieve a maximum from these elements, you should perform the following steps:
- Redundant Disk drives - organize them into a RAID configuration. RAID 1 (mirror) is the best in terms of redundancy and speed. But you loose exactly 50% of capacity. RAID 5 (parity) gives you the best trade off between capacity loss and optimal performance. When planning a RAID, look for a server that has a hardware RAID controller. The modern server operating systems can make a RAID themselves, but this way the operating system has to dedicate resources and have specific software to maintain the RAID - thus burdening the main CPU with this task
- Redundant Power Supplies - connect all power supplies of the server to power lines coming from a different circuit breaker. This will save you a lot of grief if the cleaning lady decides to connect her vacuum cleaner to an outlet connected to the same circuit breaker as the server and overloads it. If possible, connect all power supplies of the server to different Uninterruptible Power Supplies. This way, all UPS systems will help your server ride out the blackout.
- Network adapters - First, organize the network adapters to work as a failover team. This is realized with specific drivers delivered by the manufacturer, and the driver creates a virtual network adapter. The virtual network adapter is configured with the IP address of the server, and it binds to one of the physical network adapters. Should the adapter loose connectivity, the driver will bind the virtual network adapter to the other physical one, thus reestablishing connectivity. To achieve optimal solution, connect the physical network adapters to several switches which are interconnected via trunk links - thus creating one large meta-switch.
Talkback and comments are most welcome