The rise of server-based flash caching and other technologies like local performant storage pools, whether virtual storage appliances like the HP StoreVirtual VSA or VMware’s VSAN, marks a possible return to the days of the pizza box server in data centers across the world.
The pizza box server is a 1U unit that was the dominant server type before the rise of the blade format in the early nineties. Two of the drawbacks that caused its original demise were its lack of local storage capacity and its format. Then, the rise of the blade format, with the HP c7000 chassis being 10U in size and supporting sixteen half-height blades, saw an increase in density that allowed sixteen servers to serve resources in a space that previously only allowed ten. The blade format had other benefits as well, especially with the introduction of 10GbE and converged networking. Now a chassis could be serviced from just two 10GbE connections to the switching infrastructure. Taking into account remote management of the guests and chassis, an additional two Ethernet cables were required for iLO management. All are telling use cases for the blade format.
However, what about the new world of local flash caching and server-based SANs? Here, things get a little more gray. I will deal with each situation individually:
This is where local SSD-format drives are given over to the local caching of read requests, if using VMware’s local host cache, Liquidware Labs’ Flex-IO, or Atlantis Computing’s USX, or reads and writes if using PernixData’s FVP. Here, I do not think that there will be any change to the status quo. The hypervisor can be installed to an internally installed flash device or deployed to memory using Auto Deploy, and the flash devices installed in the disk bays can be utilised as flash acceleration.
Local Storage Used as Shared Storage
As already mentioned, one of the downsides of the original pizza boxes was local disk capacity. This was supplemented by the introduction of expensive storage-attached network devices. The standard half-height blade format suffers from the same limitation. This situation, however, has now changed with the 1U server format, as models like the HP DL360, Dell R420, and IBM x3250M5 all support up to eight local SAS drives. These devices can, when coupled with a local storage appliance or VSAN, form useable locally attached pools of storage distributed across a number of nodes to provide resilience.
So, What Are the Benefits?
These technologies are attractive to data center admins because, in the case of server-side caching, you remove most of the latency inherent in a SAN-based environment. Further, you increase performance of running workloads, as there is no intervening fabric or network between data at motion and compute. This means that performance increases without the addition of new datastores on your SAN or doubling SAN base cache.
What makes server-based SANs attractive to data center admins is that complexity is reduced, they gain more physical capacity, and SAN requirements are reduced. Another benefit is the ability to scale-out via the purchase of another server. This is wonderful for elasticity.
Why is this attractive to cost owners? Both flash-based caching and local server-based SAN allows them to purchase SANs for their original intended use—yes, the storage of files—and not for performance. With the heavy use of virtualization in modern data centers, there is a lot of wasted space on storage arrays due to the heavy IOPS requirements driven by hypervisor use, and profligate use of RAID1 and 10, where RAID5 would have been used previously. Further, any new storage that is needed can be utility based, perhaps using an SDS vendor on white box hardware, thereby really driving down storage costs.
We are definitely living in interesting times. If VMware gets the pricing right for VSAN technology, it can change the storage map in the same way that it changed the server map. This will then drive the sales of the pizza box server. That said, there is still a place for the blade format server, especially if flash-based caching is the chosen technology.
I can see the data center of 2016 looking very different from that which now we know and love.
Share this Article:
Latest posts by Tom Howarth (see all)
- That Was the Year That Was: 2016 - January 16, 2017
- Docker Has Been in an Acquisitive Mood Again, This Time Pulling in Infinit - January 9, 2017
- Acquisitive LANDESK Bought Out - January 5, 2017