What Next for XtremIO?

What Next for XtremIO?

Chris EvansAll-Flash Storage, Storage Hardware

The Dell EMC XtremIO team recently presented at Storage Field Day 14.  It was an opportunity to catch up on a technology I’ve been less than positive about over the years.  So has XtremIO X2 moved on?  What does the future look like for the all-flash platform?

XtremIO Background

XtremIO Inc was an Israeli start-up, founded in 2009.  EMC acquired the company in May 2012, for rumoured $430 million, before a single product had been shipped to customers.  It’s interesting to note that following a visit from Joe Tucci, two VMware execs visited the company shortly before it was acquired.  With around only $25m in investment at acquisition, XtremIO represented a great payday for the investors.  EMC certainly needed to get their money back.  As a result, XtremIO was heavily marketed.

XtremIO X1

The first XtremIO products (now renamed X1) were released to general availability in November 2013.  The architecture is based on closely coupled scale-out nodes called X-Bricks, connected by a 40Gb/s InfiniBand switch when 2 or more nodes are in a configuration.  You can read more detailed information on my blog post (here).  Some of the unique features of the platform include a proprietary RAID implementation (XDP), with only around 8% overhead.  This is achieved using 23+2 stripes, effectively encompassing all the disks on a single X-Brick.

X1 had some real shortcomings, such as the inability to extend a configuration in place.  Although the solution was technically scale-out, it couldn’t actually be scaled in the initial release.  The loss of an X-Brick was a catastrophic issue that could result in data loss because each X-Brick owned and manage a portion of the data on the platform.  So lots of extra redundancy (like battery backup and dual nodes) was put into each X-Brick configuration.  With the upgrade from XIOS 2.4 to 3.0, customers had to reload their systems as the upgrade process was destructive.  EMC changed the architecture and moved from 4KB to 8KB page sizes that couldn’t be refreshed in place.

Declining Revenues

XtremIO was touted as the fastest selling all-flash array and number 1 in the all-flash market, based on EMC figures.  However, after peaking at the end of 2015, sales of XtremIO seemed to dip significantly, as reported by The Register.  There were even rumours that the product could be discontinued.  However, this has turned out not to be the case.  In May 2017, Dell EMC announced X2, the next generation of XtremIO.

X2

XtremIO X2 has addressed some of the shortcomings of the first platform.  From a hardware perspective, there are now two configurations, the X2-S and X2-R.  The X2-S model uses smaller drives (400GB), so scales from 7.2TB to a maximum of 28.8TB per X-Brick.  The X2-R uses 1.92TB drives, with capacity from 34.5TB to 138.2TB (1.92TB drives).  Memory on each controller in an X-brick is proportional(ish) to drive size, with X2-R having 1TB per controller and X2-S 384GB per controller.  All the remaining characteristics (cores, connectivity) are the same but increased over X1 (50% more cores, twice the memory and support for 16Gb FC).

The previous X1 systems had three X-Brick capacities (10, 20 and 40TB) based on three drive sizes (400Gb, 800GB, 1.6TB).  This is now simplified in the two models.  In addition, the X2 drives are lower DPWD models, which allows XtremIO to be more cost-effective compared to other products in the market.  Dell EMC is promoting X2-S systems for higher de-duplicating workloads, such as VDI.

The X2 chassis has been redesigned somewhat, with the capability to start from 18 drives, up to a maximum capacity of 72 drives in groups of 6.  Although XDP hasn’t been redesigned, in the X2 implementation, additional drives added to a system result in multiple XDP groups being created.  Check out this post, in which I discuss XDP changes, specifically in the comments, where the XDP groups are explained.

The battery backup unit, previously used to de-stage data from memory in the case of power loss, has been replaced.  X2 systems use an NV-RAM drive and super-capacitor.  This has significantly reduced cabling but also will deliver power savings.  The change also introduces the ability to implement odd-numbered node expansion, which previously had to be done in even numbered steps.

Software

In terms of software, there are enhancements to compression, which increases the savings ratio from (typically) 4:1 to 5:1.  This is an obvious benefit for traditional platforms like databases.  One big innovation is a new feature called Write Boost.  This claims to have reduced latency by a factor of 5x.  The saving comes from having looked at the size of write I/O from data in the field. 

The XtremIO team claims that for data on XtremIO systems, 65% of all writes are for a block size of 8KB or less, so being able to reduce the overhead on small writes offers a big saving.  Write Boost works by acknowledging writes to the host at a much earlier point than occurs in X1 systems.  With Write Boost, the write I/O is acknowledged when the I/O has been processed by the node receiving the I/O and before committing to disk.  Small writes are coalesced and written to disk at a later point.

The Architect’s View

OK, XtremIO is not dead (yet).  From a development perspective, a lot of work has gone into the X2 release to address customer concerns.  Todd Toles, Field CTO at Dell EMC indicated that the company is committed to long-term support for both X1 and X2.  Presumably, new customers are already being diverted to X2, so it’s just a case of maintaining the older “legacy” X1 systems until customers transition to X2.

What can be a little confusing is the positioning of Dell EMC’s flash storage portfolio.  VDI has been mentioned a number of times now as a focus for XtremIO, but what about high-end workloads?  We touched on this in a recent Storage Unpacked podcast

XtremIO fills a gap in high-end performance where the VMAX level of resiliency or features isn’t required.  Obviously, VMAX supports mainframe (z/Series) and AS400 (i/Series), which XtremIO will never do.  So although there’s overlap, there are unique features only VMAX offers.  Todd indicated at SFD14 that the biggest use case for XtremIO is supporting traditional databases with large numbers of snapshots.  This goes back to the VDI discussion – XtremIO is price sensitive to the level of data de-duplication that can be achieved.

The improvements in XtremIO are positive and it will be interesting to see where Dell EMC takes the platform next.  Competition in the market is tough (as seen with the move by Kaminario) and by the drop in sales for XtremIO.  As ever, this is the challenge, how to remain competitive in a market that is stagnant or even shrinking.

Related Links

Comments are always welcome; please read our Comments Policy.  If you have any related links of interest, please feel free to add them as a comment for consideration.  

Copyright (c) 2009-2019 – Post #6381 – Chris M Evans, first published on https://www.architecting.it/blog, do not reproduce without permission.