Flash media is getting denser. There have been several announcements promising 50TB plus SSDs within the next six months. The distinct advantage for all-flash systems is they will be able to deliver Petabytes of capacity in just a few rack units, driving down the cost of moving to all-flash significantly and making the all-flash data center more realistic than ever. There are some downsides to these high capacity flash systems that both vendors and IT planners need to prepare.
The High Capacity Flash Networking Problems
There are essentially two networks in any storage system. The most obvious is the network that connects to the physical servers accessing the all-flash system. The second is the network internal to the all-flash array. Scale-out systems add a third network, the network that connects the various nodes of the storage cluster.
When petabytes of capacity are centralized behind a single storage system, the networks become a concern. Thanks to the high capacity and the high performance of flash, more workloads than ever will point to a single storage system entity. While each physical host will have its own connections, those hosts will be able to support more virtual machines or containers. Each host’s network capabilities may need an upgrade. The good news is current high-speed fibre channel or Ethernet connectivity should be able to support the IO load of the individual hosts.
The performance and the capacity of the flash array also means the directing of more hosts at the flash array. That means pushing the external connectivity of the all-flash array and the internal IO capability of the all-flash array to its limits. The demand may well exceed the bandwidth of current networks, plus the overhead of the networking protocols as well as SCSI translation will create additional latency.
All-Flash systems will need to embrace internal NVMe connectivity almost immediately. The problem is, at least for now, there is a price and availability premium on NVMe interconnects and NVMe SSDs. Which of course drives the price of an NVMe all-flash system up and makes it less appealing. One solution is to create a small NVMe Flash tier to handle inbound write IO and integrate it with a high-capacity SAS flash tier, using high-capacity drives. In this design the high capacity tier is primarily used to store data that is not immediately (within the last 24 hours) active. The design is a new hybrid, that instead of tiering between flash and HDD, tiers between high-performance flash and high-capacity flash.
High capacity flash systems promise to reduce the cost of the all-flash data center while making the aspiration more realistic. IT planners are going to need to take steps to make sure that both the internal and the external network do not bottleneck performance as these high capacity drives become more prevalent.
The network is just one area of concern that IT planners need to recognize. In addition, they need to worry about fault domains and how to not buy so much flash capacity that half of their investment is idle. In our on demand webinar, “All-Flash in 2018 – What You Need to Know”, Storage Switzerland and Tegile Systems discuss how to plan your 2018 flash deployments.
Attendees will learn:
- How NVMe Flash and NVMe over Fabrics promise to deliver unprecedented levels of performance and low latency.
- How Ultra High-Density flash pledges to open up new opportunities like flash for big data, secondary storage and even archive.
- The role of hard disk and tape in a flash dominant data center.