>>>We want to reevaluate our server strategy in order to make it more robust.
>>>
>>>We now have two servers (2003 and 2008) and a PC where we store extra copies of some directories. The servers are used mostly for serving files, with SQL Server and Backup Assist as really the only two server apps.
>>>
>>>I would appreciate any advice and warnings.
>>
>>What are your current concerns?
>>
>>What new functions or capabilities do you want to be able to support in the future?
>
>Hi Al. In the past we have had server disk failures. Since we do make nightly bakcups with Backup Assist we were able to set up a different machine to be server while server machine was fixed, but that is not good enough.
>
>We now plan to add a server and want to structure things in a more robust manner, say by making nightly backups to another server that is practically ready to go. We want a cost efective arrangement, Instant on is not required, but the process should be reliable, quicker and less stressful. Suggestions such as keeping backup machine in a separate room are welcome.

Your first line of defense against disk failures is to use a RAID array that has redundancy. At the very least, the system volume of any server should be at least RAID1. That way, if one disk fails, you get a warning, but in the meantine the server continues to run as if nothing has happened. You then have time to obtain a replacement drive. You take the server down briefly to replace the failed drive, and when you restart it, with most modern RAID controllers, the RAID array rebuilds itself automatically.

My main dev workstation has 2x 1TB drives in RAID1. Since February 2010 I've had 3 drive failures, but I've suffered no data loss and only minutes of downtime while replacing drives. The automatic array rebuild does take around 9 or 10 hours, but I can continue to use the system while it's rebuilding (although disk performance is a bit less).

RAID1 is inexpensive. In my case it's just the cost of the extra hard drive, I'm using the Intel ICH10R controller that's built into my motherboard. Even if you need to purchase a separate RAID controller, they're currently not that expensive.

You can also consider so-called hot-swap RAID. This is special disk backplane hardware that allows removal of a failed drive, and replacement with a new unit, without taking down the server. If you have requirements for high availability this is an (albeit expensive) option.

Another semi-radical idea is to use a single solid-state disk (SSD) for your system hard drive. Because there are no mechanical moving parts, they are inherently more reliable than conventional magnetic hard drives (HDs). While overall reliability may not be as good as RAID1 HDs, it's probably better than a single HD. Performance can be extremely good - some server-class SSDs have performance as good as, or better than some multi-drive arrays of HDs.

The drawbacks to SSDs as server drives:

- At this time they don't have a long track record of reliability
- You should not use them in simple RAID arrays (e.g. RAID1) because current controllers don't fully support them
- They are relatively low capacity, and have a high cost/GB compared to magnetic

Some configurations use a hybrid, such as a single SSD for the system drive, and RAIDed HDs for other volumes.

If you're able to use your existing backup to establish another server, that's good, that means you already have an adequate (and tested!) disaster recovery plan in place. That is good in case of fire, theft etc.

If you'd prefer to stick with a single server, you can get good reliability if you ensure you have server-grade equipment: motherboard, RAM, disk subsystem, power supply etc. Make sure the server has a good UPS, and it's adequately cooled and kept free of dust accumulation.

For larger environments where high availability is important, you can consider:

- Clustered servers - the same idea as RAID1 but at the level of the complete server computer.

- Virtualization. You can establish two or more powerful servers running a virtual machine manager (VMM) such as VMWare or Hyper-V, and run virtual server instances on them. Your current server installations can be migrated to virtual via a physical-to-virtual (P2V) conversion process. With modern VMMs you can move running VMs from one physical server to another. You can also cluster server VMs and set up automatic failover.