“The best IO is the one you don’t have to do.”
That’s famous advice from American computer architect and high-tech entrepreneur Gene Amdahl, and it’s words to live by in today’s IO-intensive virtualized infrastructure.
A previous blog in this series detailed the data problem: random IO exposing IOPS as one of the most expensive resources in the data center due to the IO density issue with HDDs. Capacity and performance increases have not been in lock step, resulting in the decline of IOPS per GB.
The data problem has been the catalyst for innovation, such as SSD (SSD arrays, SSD caching, SSD drives in servers, SSD drives in storage arrays, etc.). However, isn’t SSD just treating the symptom? Shouldn’t Gene Amdahl’s advice be considered? Think about the reduction in data center complexity if the myriad of devices devoted to optimizing storage could go away.
What if, instead, IO could be deduplicated?
That is, what if subsequent writes to HDD for data segments that have already been written to HDD were eliminated? That’s the definition of deduplication. Deduplication identifies redundancy and eliminates that redundancy. Less data is stored, saving storage capacity. However, more importantly, redundant reads and writes are eliminated from/to disk, deduplicating IO.
Further, if the data is maintained in its deduplicated state, then the IO that would have occurred as data moved between different devices devoted to optimizing storage would be eliminated. The data would not have to be rehydrated (or un-deduplicatied), moved and then deduplicated again at a different device—a process that increases read/write IO.
SimpliVity’s OmniStack Data Virtualization Platform is the core technology in its hyperconverged infrastructure that performs inline deduplication, compression and optimization. These processes occur across all storage tiers within the hyperconverged infrastructure, including DRAM, flash/SSD and HDD).
But wait, can’t inline deduplication and compression negatively impact performance?
Yes, it can. That’s why many storage vendors over-provision for performance with more RAM or more processing power, adding to costs and creating inefficiency with those resources. For hyperconverged infrastructure, adding more processors or increasing core count can greatly impact TCO since applications are more often licensed based on core count.
SimpliVity takes a different approach to inline deduplication in its OmniStack-powered hyperconverged infrastructure—one that enables high performance with no impact to latency. The breakthrough technology SimpliVity’s Data Virtualization Platform includes SimpliVity’s OmniStack Accelerator Card, a PCIe module that contains a field-programmable gate array (FPGA) and NVRAM to offload processing of the writes to HDD in real time. The OmniStack Accelerator Card is a standard component of SimpliVity’s hyperconverged infrastructure.
The next blog will deliver greater detail about the OmniStack Accelerator Card.