Server "Deployment" with Flash Snaps

While the number of bare metal servers (those running non-hypervisor) host operating system is currently declining, there remain significant challenges to system cloning including operational complexity and security risks.

During USB initiated cloning or installations there are risks from the use of USB media itself, the contradiction of USB booting which explicit bypasses UEFI mechanisms, potential human errors, the potential for security breaches in the clone bootstrap OS and application itself, and numerous other disadvantages. Network boots to clone are far less susceptible to media breaches but still involve a number of configuration choices and involve the potential risks of an automated installation of an undesired server.

Using iscsi remote boot can alleviate these challenges and improve the delivery process. Using a flash platform you can snap and re-deploy (re-name) an OS including hypervisors (ESXi for VMware, KVM, Hyper-V, etc.). Although the snap and volume create doesn't allow customization, it is easy to set aside a block of IPs and keep volumes with those (management) IP's ready for deployment, snap, connect the volume, and then instead of bootstraping, boot to the desired OS and change the management IP.

In addition to the speed of booting (relative to PXE or vs. cloning), the process is perfectly error free excluding only the manual steps of changing the IP address. Other benefits of a iscsi remote boot include the potential to use CHAP to eliminate the threat of rouge hardware, although that does require an additional step on the host prior to booting from iscsi. Another benefit is the ability to deploy upgrades by renaming and re-connecting volumes (rather than actually running software upgrades).

With container based architectures we are likely to see a resurgence in bare-metal OS deployments. Although iscsi remote boot offers a solution for the OS deployments, it could be a challenge for the creation of container overlay file systems. Certainly some vendors are building software to provide the file system container overlay for block (iscsi in this case) storage, but NAS based approaches offer a potentially more natural fit to the logic architectural of containers. That is a per container file system which operates using NAS (in the backend).

Whether the technology will shift entirely to NAS or become hybrid (boot from iscsi for host OS but run containers on NAS) remains to be seen. Either way NAS is likely to return from its "low perfomance" corner of the data center to a central position. Where block may typically be 75% to 85% today with 25% to 15% NAS, its seems clear that will reverse with containers to more like 70-80% NAS and 20% to 30% block, with the likely continued decrease in block overtime.

While flash hardware gets most of the credit, the hidden heroes are Ethernet and PCIe. Both in price and performance Ethernet can now perform -- at speed easily better than SAS or SATA, IO throughput of 10Gbps (at least) if not 40Gbps to 100Gbps (depending on the network interface cards used on the servers and the storage arrays, the number of PCIe lanes used, but also on the hash mechanisms of multipath forwarding algorithm(s) used for iscsi forwarding. While the most reliable and secure implementations remain a bit tricky, the trade off in power, space, efficiency, easy of use, and cost -- clearly demonstrate that by ALL of these measures flash over 10G/40G/100G is the only viable scale option.

Still block has years of life left in it and the benefits of a flash block platform are immediate today. Automated provisioning, elasticity of volumes, use of Ethernet based transport, IO performance, and not least power and space density make it worth the transition even if you are not ready for NAS.