Diskless Server Provisioning

If you were to take a walk around some of the worlds most advanced datacentres, perhaps those that are run by the likes of Microsoft or Google - you would notice something rather different about the sorts of servers they use.

Most of them do not contain hard disks.

In this world of 'cloud computing' where the underlying files and operating system for a server is not tied to any particular geographic location, naturally the ability to unshackle the server hardware from the server software, is key.

For high end enterprise type implementations - this has traditionally be achieved with the use of various SANs (Storage Area Networks) which provide the networked storage for the server. However, in order to 'talk' to this storage, servers typically need a Host Bus Adapter or HBA. These HBAs would allow a diskless server to connect to the assigned hard disk, which resided on the SAN - perhaps miles away. 

The benefits of separating hardware and software like this are numerous. 

For one, the severs themselves do not need expensive disks or disk controllers. If the SAN the server connected to resided in a high availability cluster - you would be very unlikely to experience a server hard disk failure - because the servers don't have hard disks! SAN systems have advanced block snapshotting technology that mean that if you had a problem with a server volume, (corrupt data, virus, or wholesale accidental deletion) even a volume that contained many terabytes of data could be restored from the previous OK snapshot - in milliseconds. Then there are the power savings of removing hard disks from servers. Many servers use very inefficient levels of RAID (such as RAID 1) where all data from one disk is 'mirrored' to another - effectively wasting an entire disk. Without such disks or raid level on the server - there is a much higher level of storage efficiency. Think of the I/O that a server would be able to achieve if it were connected to a large IO rich SAN cluster - rather than it's own local disk subsystem. As volumes in SANs are 'thin provisioned' only storage this is used is provisioned, rather than the entire amount of disk space allocated to a server. So for example, your SAN would 'tell' your server that it has a 1TB volume, and this is what would show in the server. But in real terms, if the server only had 100GB of data on the volume - that is the amount of data which would actually be used on the SAN. 

So there is no argument about whether or not SANs are a good idea. The only argument is how to plan, afford, and deploy them.

For one, HBAs are expensive. Perhaps there will be hardware problems with fitting the cards to your servers? Maybe your servers don't have the right sort of expansion slots? You might be looking at buying completely new server hardware. When the new servers come in, now you have to configure each HBA, on the card, so managing 100's of them could be a challenge.

So this is where Flex Management Server comes in. Flex Management Server enables you to turn a standard server into a boot from SAN server - as long as it contains a gigabit network card, and the ability to PXE boot - which most gigabit network cards do. In many cases, additional hardware is not required to allow you to separate your hardware from your software. 

If you untie your server hardware from the software, if you are unlucky enough to suffer hardware failure, it's simple to power up some spare server hardware 'bare metal' and enable this spare server to take over. With remote power control and KVM over IP systems - this is all possible to do remotely. Unlike the HBA approach, the linkage between the SAN and the hardware is all performed in a centralised user interface, in which you can simply drag and drop the appropriate volumes to appropriate server hardware.

It is true that in a virtualised server environment, at the guest OS level, similar actions can be performed. However, for high performance - where this is no 'hypervisor hit'  - nor is there any virtualised 'expense' - a Flex based boot from SAN architecture will always win. 

That said, with the advent of HyperV from Microsoft, it's quite straight forward to boot your HyperV server from SAN - giving you the ability to replace your HyperV server hardware remotely - if a hardware issue were to occur. This is something that can be done in minutes. The alternative is to run two HyperV servers, and these would need to be of Enterprise edition type - a hugely more expensive setup. 

Flex Management Server makes the relationship between your server hardware, and the software that runs on a totally flexible.