Not all hyperconverged systems are created equal. If 2015 was the year of hyperconvergence, then 2016 appears to be the year when the term becomes completely overloaded. Whenever a market segment enters a phase of extreme growth, it attracts a crowd of new products that attempt to market and brand themselves into that space. Unfortunately, this can create a tremendous amount of confusion in the market and make it much more difficult for IT buyers to decipher the market.
One key dimension of a hyperconverged system is that it provides all of the data storage for virtual machines. Storing data is not something to be taken lightly. There is nothing more important than data in IT. A server can be replaced. A network can be replaced. Data cannot be replaced. This is why SimpliVity is so focused on data protection. Every OmniStack system, whether it is the enterprise grade server we OEM or a server from one of our partners, Cisco and Lenovo, protects the data internally with RAID. This protects the data in the event that a HDD or SSD fails in a system. SimpliVity’s architecture also leverages RAIN since the data is also saved to two separate systems to protect from any type of node failure within the environment. SimpliVity applies these multiple levels of protection because customer data is critical.
Unfortunately, not all vendors recommend the same level of data protection. Some systems in the market use the technique of writing two copies of the data, each one to a different node. This is commonly referred to as Replication Factor 2 or RF2. In systems that use this level of data protection and distribute the data, two HDDs in a cluster likely contain data from many different VMs. In an RF2 environment, the failure of two HDDs in two different nodes can cause catastrophic data loss. This is why it is so critical to ask any hyperconverged vendor you speak with how they are protecting your mission critical data. Make sure to find out how many nodes are required to minimize risk of data loss and what failure scenarios the vendor can handle vs. not.
Vendors claiming to offer hyperconverged platforms have also started to push the boundaries of what it means to persist a write. In a SimpliVity environment, every system has an OmniStack Accelerator Card. This accelerator offloads all of the heavy lifting of data efficiency processing (deduplication, compression, and optimization) from the Intel CPUs in order to leave as many resources available for the business applications as possible. It also accelerates writes by saving them to a DRAM buffer on the card. This DRAM buffer is protected by a bank of super capacitors that flush the DRAM to flash in the event of a power loss. In order to provide maximum data protection, writes from a guest VM are not acknowledged until the write has been saved into the OmniStack Accelerator Card on two different systems.
There are several ways that other vendors are putting customer data at risk. One is by acknowledging writes as soon as they are staged to host DRAM on two servers. This is incredibly dangerous, as a simple power loss on two nodes will trigger data loss. The guest VM will see the data as committed, but anything sitting in DRAM will be lost. This is a perfect example of what happens when a platform that was not designed to host mission critical data is forced into a datacenter. Customer data is placed at risk. One of my colleagues describes this as “eventually persistent” and I think it is very fitting. I also find it to be a terrifying concept for data protection.
The moral of the story is to ask questions about the hyperconverged platforms you are investigating. SimpliVity designed the OmniStack Data Virtualization Platform from the ground up to protect mission critical customer data. Unfortunately, not every vendor takes data protection as seriously. Learn more about SimpliVity’s architecture.