Close, but No Cigar!
By Jeff Feierfeil Director of Product Management, Nimble Storage
That phrase dates back to the mid-20th century when fairground booths gave out cigars as prizes. In the carnival that is today’s data storage market, it is expected that vendors will leapfrog one another in spec sheet comparisons. But it’s odd to see a vendor make a big deal about products that don’t even bring them up to par with their competitors. At least that’s the impression we got from Pure Storage’s recent product announcements.
We applaud Pure’s intention (planned for “Q4 of calendar 2016”) to ship “Always on QoS”. We agree that most customers would like their arrays to automatically prevent “noisy neighbors” from starving the performance of other applications on an array. This is why we started shipping “Auto QoS” early this year. With an eye to our quickly growing Service Provider customer base, we have also been shipping a tenant abstraction that will make it much easier for Service Providers to offer differentiated QoS levels per tenant in the future (by setting limits per tenant or application). It appears that when Pure does get around to matching Nimble on QoS limits (planned for a future release), Service Providers will be forced to manage limits per volume without the benefit of such a tenant abstraction.
Pure also made significant improvements to maximum array capacities. At the time we started shipping our All Flash arrays in March of this year, there was a huge gap in maximum raw flash capacity of a single array (Nimble: 553TB, Pure 136TB). Pure has now narrowed the array gap, but still lags in two areas:
- Nimble supports both scale-up and scale-out, achieving a more than 8PB of all flash capacity in a single Pool. Pure only supports scale-up.
- Nimble’s more efficient flash filesystem delivers an additional 20% usable capacity for a given amount of raw flash capacity (usable is the amount of capacity available after taking into account RAID, over-provisioning and other system overheads).
It’s also interesting to compare the very different approaches both companies have taken to achieve array scalability. Being a newer generation of AFA, we focused on building an architecture suited to large flash capacities and always-on data services. This resulted in a highly CPU and memory-efficient architecture. Comparing the flagship product from both companies: Pure requires ~4.6x more memory, to deliver 30% less usable capacity and one third the number of snapshots. Not to mention that while Nimble’s inline deduplication never shuts off, Pure has to disable inline deduplication to keep up with heavy write workloads:
Now The Important Part
Although Pure has caught up with Nimble in some array specifications, what hasn’t changed are the things that customers really care about.
Reliability and Resiliency
Arguably the most important aspect of any storage platform is reliability and its ability to withstand and correct failures without compromising data or data services. This is why Nimble uses Triple+ Parity RAID – three parity drives per RAID group plus intra-drive parity in every SSD. This allows a Nimble array to continue working even after the loss of three entire SSDs and sector failures in every other SSD. By contrast, Pure uses what we would term Double+ Parity (two parity drives plus intra-drive parity). From a hardware perspective, if an SSD fails, Nimble allows it to be replaced without compromising any other SSDs or RAID groups. In contrast, Pure forces two SSDs to be replaced at a time (they keep two SSDs, belonging to two different RAID groups, in each “Flash Module”).
InfoSight Predictive Analytics
More than eight years ago, we designed our arrays to leverage a network effect for machine learning. We set the goal that if one customer encounters an issue, small or large, we would use data science to learn the root cause and then automatically inoculate every other system from hitting the same issue. We started collecting data sensors from just our arrays, but recognizing that the majority of infrastructure issues that affect applications are not caused by storage, extended our collection all the way up the infrastructure stack to the virtual machine. Today, we collect and correlate more than 250 billion sensor data points a day from the infrastructure stacks across our installed base of more than 9,000 customers. This has some profound benefits for our customers:
- InfoSight automatically opens and resolves 86% of issues before a customer even knows there is a problem. 54% of the issues InfoSight resolves are not even caused by the storage (networks, hosts, VMs, OS configuration, etc.).
- For the small number of issues that require a call to support, we are able to rapidly resolve the problem down to root cause thanks to the telemetry already collected by InfoSight and the fact that our support centers are staffed with only the industry’s best Level-3 support engineers.
- The InfoSight SaaS portal automates planning tasks by using machine learning to accurately predict future needs and bottlenecks. The portal also provides executive dashboards and cross-stack visualizations giving users just the information needed to run their environments optimally.
Unified Flash Fabric
Nimble has a single unified architecture that spans All Flash and Adaptive Flash (hybrid) arrays. This allows users to leverage All Flash for primary workloads and then replicate to Adaptive Flash to achieve backup, disaster recovery and archival at a third of the cost. It also allows customers to implement a storage solution where performance sensitive workloads use All Flash and applications that are more cost-sensitive use Adaptive Flash, but manage them as one.
Why No Cigar for Pure?
The storage market is going through a massive transition as enterprises implement All Flash and move away from legacy storage solutions. The pace of this transition is so rapid that customers are still trying to determine what is most important when they select an All Flash vendor. While the array’s specifications are clearly important, customers are quickly realizing that almost any All Flash array will deliver performance that is massively faster than the solution it replaces.
This is why Nimble is taking a two-pronged approach to All Flash innovation. We are committed to always building arrays that deliver leading-edge performance and capacity. But far more importantly, we are focusing on the differentiation provided by InfoSight – predicting and preventing issues across a customer’s infrastructure to fundamentally transform how applications perform.
- Jeff Feierfeil