The press releases are out there but they only cover the XtremIO announcement at a high level. It’s time for a deeper dive. At Dell EMC World, Storage Switzerland spent quality time on the XtremIO X2 update. It promises increased density and efficiency but it may present an upgrade challenge to existing customers.
When EMC bought XtremIO it placed a big bet on an all-flash storage system that had not yet shipped. The company guided XtremIO to maturity and by all indications the bet paid off. XtremIO is considered to be one of the share leaders in the all-flash market. X2 should continue that success, but EMC does has to navigate the upgrade process for its current customer base.
The X2 builds on the XtremIO X1 architecture which leverages intelligent use of metadata to optimize flash operations. This intelligence powers data services that enable in-line deduplication and compression as well as an impressive integrated copy data management capability.
What’s in XtremIO X2?
The new release provides new hardware and software. From a hardware perspective the system has more processing cores and a controller head that is 4U instead of 6U. The system also ships with higher capacity flash drives which deliver greater density.
The All-Flash Scaling Problem
The decision between scale-up vs. scale-out storage is of great concern in the all-flash market. Because of the performance of flash and its ability to deliver low impact, data efficiency features many organizations can get more than enough performance and capacity in a single system. But there is always the potential that at some point the storage system will hit either performance or capacity limits and the customer will have wished that they bought a scale-out.
Scale-out systems on the other hand, often require a three node starting point and each upgrade step is in the form of a fully populated node. So while the scale-out system can grow with the customer, there are points in that growth where the system is massively over-sized for the task at hand.
There are a few systems on the market that partly solve this issue by starting as a scale-up system and then can scale-out through the addition of nodes. This scaling method is often called scale right. The problem is that once the system reverts to scale-out “mode” it still suffers with the same points over being overkill as nodes are added. Essentially the scale-out architecture is out of balance when a new node is added to the cluster, it offers too much additional capacity and too much additional performance.
X2 Scales Righter
XtermIO X2 meets the scaling challenge in a couple of ways. XtremIO X1 has the ability to scale right. Meaning that an organization can start with one node and then add additional nodes as it needs to. But it too suffers from moments of being too much for the customer as the organization adds nodes. X2 solves this problem. First, the initial node scales to much higher capacity and offers greater performance. It can hold 72 drives per node, which Dell EMC calls an xBrick. The xBrick uses 1.92TB SSD drives and now have a raw capacity of 138TB per 4U node.
The big difference though is that XtremIO can scale more incrementally than in the past. Each xBrick starts with 16 drives and then can add capacity in increments of six. Meaning the customer will likely not face the “out of balance” problem. By the time customers fill up the node with drives because of capacity demands, it is likely they will also need the performance bump that the next node will deliver.
The impact of 1.92TB SSDs has a twofold impact on cost. First, the cost per GB of a 1.92TB drive is less expensive than a lower capacity drive. Second, with the exception of some fringe performance situations, the time before having to add and pay for an additional node is increased. In other words each node lasts a little longer. The incremental nature of the addition of drives per node makes the use of such high capacity SSDs more practical.
X2 Data Efficiency
The above numbers are raw capacities. But assuming XtremIO’s claim of a 25 percent improvement in data efficiency is accurate, the system should now deliver a 6:1 data efficiency ratio instead of the 5:1 ratio that the industry often leans on. A 6:1 ratio means the per node capacity of an xBrick could be 830TBs. For many organizations the xBrick will meet all of their capacity requirements.
XtremIO gets the 25 percent gain in data efficiency by performing a function that they call intelligent data packing. The technique essentially makes sure that there is less wasted space on the SSD by writing data more contiguously.
Part of the challenge in describing the new XtremIO release is some changes impact more than one area of the offering. Data packing is a good example. It seems that data packing is a side benefit of X2’s use of NVRAM to improve performance. Since data is secured in NVRAM, it can be coalesced and written more intelligently. Not only should this improve data efficiency it should also improve drive reliability which also partly is what enables XtremIO to use lower write per day, but higher capacity 1.92TB drives.
The NVRAM’s primary purpose is to improve write performance, and is especially evident in small write IO, which Dell EMC claims is more than 60 percent of the traffic their customers generate. In X1, without NVRAM, data is written directly to flash. That means on an update, data has to be read from flash into ram, updated and written back to flash, before an acknowledgement is sent to the application. In X2, data is written to two NVRAM modules, and then an acknowledgement is sent to the application. Later in the background that data is written to flash. Again, as part of this process the write is also coalesced and now “packed.”
Dell EMC claims that the inclusion of the NVRAM modules should reduce application latency by 80 percent, improve IOPS performance by 26 percent and provide a 200 percent improvement on copy operations.
After the initial release, X2 will also get native replication. The replication will leverage the system’s metadata intelligence to only send blocks are are not only unique to the system but also unique to the target. As a result, overall, the amount of data being sent across the WAN should be the smallest amount possible.
Improved Data Copy
X2 will initially ship with similar data copy capabilities that Dell EMC calls, “intelligent copy data management” (iCDM). After the initial X2 release, Dell EMC will add QoS capabilities to the solution. Storage administrators will be able to set maximum IOPS and bandwidth thresholds. X2 will also have a burst mode setting where the volume will be allowed to exceed maximum thresholds for a period of time.
The X2 Challenge
X2 is an impressive upgrade for the XtremIO family. But it does have one flaw that I think Dell EMC’s competitors will pounce on. The X1 hardware can not coexist with the X2 hardware and software. X2 hardware will be a separate cluster, but IT can manage both clusters from a common interface. While the competition will claim Dell EMC broke the first rule of scale-out designs, there are specific hardware requirements of the new X2 software that the old hardware can’t accommodate.
While managing nodes of different capacities and drive types should be possible, I’m not sure how a developer accommodates for lacking hardware, like for example NVRAM. How is the storage software supposed to navigate around the lack of NVRAM in some nodes when it is expecting it to be there?
Certainly there are tools both in the Dell EMC family and externally that can migrate data between the platforms. The challenge will be what to do with X1 customers that want to switch platforms. Essentially switching becomes a business problem more so than a technology problem, something that Dell EMC have experience navigating.
The XtremIO X2 is an impressive product. It builds on the X1 foundation and solves a lot of practical problems first generation scale-out storage systems encountered, especially in terms of how the system scales. Most all-flash arrays periodically claim a performance improvement mostly because they switch to hardware with faster CPUs. Dell EMC XtremIO is doing that and more by adding NVRAM and other intelligence to the software.
There will be concerns about the X1 to X2 transition, but that only impacts existing customers and is something the Dell EMC and their customers will have to work through. For first time all-flash customers and non EMC all-flash customers looking to upgrade, X2 is very compelling.