Doing Solid-State The Right Way
It’s safe to say that solid-state storage is the most expensive component in most storage systems. As such, you’d expect vendors to be mindful of this and make it as efficient as possible. So it puzzles me to see so many vendors treating solid-state like inexpensive spinning disk. For example, stuffing solid-state into any available drive bay and then applying RAID. It’s just plain inefficient. You end up paying dearly for capacity and performance that is never be used.
This waste is compounded when all data is written to solid-state regardless of its performance needs. Why is this wasteful? This may be a shock to most, but in general purpose storage systems only 3% to 5% of data is active! This means that 95%-97% of data is idle and rarely accessed. Why would you store data that’s hardly ever accessed on the most expensive type of storage?
NexGen has designed a hybrid storage system that offers the best balance of performance and capacity, geared to heterogeneous workloads where 3% - 5% of data is active and the rest is not. Our intelligent datapath maximizes your solid-state investment, and eliminates the solid-state waste associated with other vendors’ solutions. Here’s how…
First, we ensure that all writes going to solid-state are highly available. To accomplish this, data is first written to solid-state on one of the two active/active storage processors, and then it is immediately mirrored to solid-state on the other storage processor. At which point, data is highly available, and the write is acknowledged to the host (Figure 1).
We recognize solid-state is an expensive storage tier, so to minimize waste and maximize its goodness, the mirror is only temporary. As soon the write is acknowledged to the host, the mirrored copy is immediately migrated to the NL-SAS tier. At this point we have the original copy on solid-state and a redundant copy with dual parity on disk (Figure 2). Now, more s-state performance is available for both read and write workloads.
3. At the risk of sounding like an infomercial, I will. “Wait! There’s more!” To make solid-state as efficient as possible, NexGen is always inspecting data on solid-state to ensure that the data actually requires solid-state performance. If it hasn’t been accessed in a while, NexGen will evict it from solid-state completely. At which point the data will be available on NL-SAS, and space will be freed up on solid-state for data that would benefit from it (Figure 3).
We’ve sized our system to match the ratio of active to inactive data found in general purpose storage systems. Now, data that needs low latency for reads and writes is stored on solid-state and inactive data is stored on slower but significantly less expensive disk. This approach ensures the lowest $/GB while delivering solid-state performance.