Treat Storage as Black Boxes to Optimize Infrastructure Designs
Gartner, with the publication of their 2019 Magic Quadrant for Primary Storage, which includes both Solid-State Arrays (a.k.a. AFAs) and General Purpose Disk Arrays (a.k.a. hybrid arrays), has told the market that competition between AFAs and GPDAs will continue for the foreseeable future. We agree with that assessment because our InfiniBox is proving its competitiveness consistently against AFAs by winning technical and financial bakeoffs. More specifically, InfiniBox has inherent advantages in scale and cost relative to AFAs and is competitive in performance, usable availability, and ease-of-use. We also believe that affordable scalability is particularly important when designing infrastructure refreshes to support storage consolidation, data analytics, and AI/ML/DL applications: the things that create new capabilities and competitive advantage.
Setting aside the AFA versus hybrid array debate for a moment, consider the following question - “How would you choose between two black-box storage arrays built with mysterious unknown technologies?” The answer to that question is your roadmap to building a better storage infrastructure because it begins with defining your organization’s existing application portfolios storage needs, not esoteric discussions about architectures, technologies, and algorithms or cloud-based analytics.
These storage needs ultimately translate into characterizations of your replacement arrays and at least a mental RFP that specifies their scale, performance, availability, replication capabilities, ease-of-use, ecosystem support, post-sales support effectiveness, environmental demands, and financial considerations.
Table 1 compares AFAs, traditional hybrids, and InfiniBox across these and other measures of product attractiveness without referencing media or any other technologies, highlighting the areas where Infinidat has delivered disruptive improvements.
Table 1 - Storage Array Product Attractiveness
Perhaps the most surprising positioning statements in Table 1 are InfiniBox claims of performance and environmental footprint parity with AFAs because we all “KNOW” that SSDs are faster than HDDs and that they consume less power on a TB per Watt basis than HDDs. The problem with this analysis is that it is a media-level analysis, not an array-level analysis, that ignores:
- InfiniBox services 100% of write I/Os out of DRAM. That is 4 orders of magnitude faster (10,000X) than NAND Flash. Reads are serviced out of DRAM, SSD read cache or HDDs on cache misses. With Neural Cache AI/ML typically delivering 90%+ cache hit ratios, the problem with media-based analysis is self-evident.
- High-capacity 7,200 RPM HDDs TB/W ratios are competitive with SSDs because they consume much less power than 10K RPM and 15K RPM HDDs. For the curious, the motor accounts for most of the power consumed by an HDD and that goes down as the square of the head’s velocity relative to the platter. See Table 2 below.
Table 2 - HDD vs SSD TB/W
- The multiple AFAs needed to match InfiniBox scale further advantages Infinibox because every AFA or AFA storage node has at least two controllers and redundant power supplies, cooling fans, and converged network adaptors all of which consume real estate and power.
What are AFA’s persistent advantages?
Pre-InfiniBox this was an easy question to answer: consistent sub-millisecond response times, much better $/IOP price-performance than hybrid arrays, unbeatable ease-of-use, more reliability, and a much smaller environmental footprint than hybrid arrays with equivalent I/O capabilities. Post InfiniBox, AFA vendors are left with four relatively weak arguments.
- “My sub-millisecond response times are lower than InfiniBox’s sub-millisecond response times.”
This is an argument we would certainly contest. We would also ask, is a couple 1/10s of a millisecond in response times worth more than doubling or tripling your storage capacity without increasing your storage budget? Given that most OLTP applications are quite happy with sub-millisecond response times and that the use of archiving complicates storage infrastructures and operations, I suspect not!
- “Our scale-out architecture allows us to start small and grow to very large configurations.”
This is mostly an academic argument because typical CAGRs and maintenance costs hold growth to no more 2X to 3X times the initial capacity of the array over a 5-year period. Other reasons include node failures creating performance problems and 3-tier infrastructures’ superior economics and manageability overtaking HCI at scale.
- “InfiniBox’s use of HDDs for persistent storage makes it more sensitive to workloads with very high write content (low r/w ratios) than AFAs.”
Possibly true with atypical workloads like real-time data collection from nuclear experiments, but utterly irrelevant because we benchmark customer workloads as part of our sales cycle so there are no performance surprises post-installation.
- “Our TB scale and use of NAND Flash make us a better solution at the edge where environmental advantages matter.”
We agree and would note that the edge is not our target market because our packaging creates a physically larger entry point than a two-node cluster with direct-attached storage.
The following rationales support the positioning statements in Table 1.
Infinidat’s decision to target the high-end of the storage market and differentiate itself from AFA vendors with scale and disruptive economics resulted in a multi-PB scale, self-managing, mixed-media architecture that delivers sub-millisecond response times with HDD economics. The F6300, our flagship array, can be configured with up to 4 PB of usable capacity which translates to ~10 PB of effective capacity. By comparison, Pure Storage’s newly announced “capacity model”, the FlashArray//C60-1390, can be configured with up to 1.39 PB of raw capacity or up to 5.2 PB of effective capacity. A fully configured PowerMax 8000 has a “Typical Effective Max Capacity with Data Reduction” of 4 PB. In other words, the biggest AFAs have approximately half the capacity of an InfiniBox, and InfiniBox performance does not fall off as the array fills up.
Any discussion about performance should always be caveated with the admonition that “your mileage may vary”. That said, InfiniBox is a self-managing storage array that delivers:
- Consistent sub-msec response times
- DRAM/flash cache hit ratios in the 90+% range
- Up to 1.4M FC/iSCSI IOPS and 2M NFS IOPS
- Up to 25 GB/s of host visible bandwidth
Figure 1 shows the cache hit ratio experienced by an InfiniBox retail customer over a one-week period beginning on “Black Friday”, the busiest shopping days of the year. Per Figure 1, shows 90%+ cache hit ratios which reduces average host visible InfiniBox latencies by an order of magnitude.
Figure 1 - InfiniBox cache hit ratio at high load
While some Infinidat competitors have published higher “hero” numbers that give them plausible bragging rights, we model our customers’ workloads to ensure that their InfiniBox has not been under-configured and will perform as promised when put into production. And yes, we will provide our customers on a case-by-case basis with performance guarantees that include meaningful remedies.
While InfiniBox’s exceptional performance may create a cognitive dissidence problem for users committed to AFAs, it does provide users not needing more than 1.4M FC/iSCSI IOPS or 2M NFS IOPS and 25 GB/s of bandwidth the ability to double or triple their capacity without increasing their storage budget.
InfiniBox delivers 100% data availability, but more important than that is the inherent performance advantage the active/active/active architecture provides in the presence of hardware failures. Users that follow the best practice of leaving themselves 25% to 30% of headroom to cope with unforeseen performance problems and to give themselves time when negotiating infrastructure refreshes should be able to meet service level agreements, even in the presence of a controller failure.
InfiniBox’s all-inclusive software pricing model reflects the design of InfiniBox software. Metadata and data flows were designed from the beginning to support a rich set of local and remote replication capabilities. The result:
- InfiniSnap – space-efficient snapshots that have zero impact on performance, can be made r/w mountable, has no requirement to pre-provision snap capacity, and has consistency group support.
- Infinibox Async Replication - is a snapshot-based solution that protects the data by replicating it to a remote site without adding latency to the host I/Os. The replication of the data is done asynchronously with a minimal 4 secs RPO (Recovery Point Objective).
- Infinibox Sync Replication - is the synchronous replication solution that protects the data with zero RPO. Every write to the local system is immediately written to a copy of the dataset on the remote system before returning to the host.
- Infinibox Active-Active Replication - is a synchronous replication solution that protects the data with zero RPO and zero RTO (Recovery Time Objective). The data of the replicated datasets can be accessed by the host on both systems at the same time.
- InfiniSync –is a highly resilient zero-data-loss solution that enables a simple two-data-center topology for an RPO of Zero over asynchronous replication (Recovery Point Objective) providing complete data integrity and application consistency at any distance.
Ease of Use
Customers and Gartner Peer Insights both rate InfiniBox as very easy to manage because InfiniBox manages backend data placement and uses QoS to ensure that service level objectives are consistently met without the need for tuning, episodic garbage collection, or archiving to keep the cost of storing cold data affordable. Cloud-based monitoring and reporting simplify software change control, and performance modeling practically eliminates unpleasant surprises. InfiniBox scale and performance further simplify management by enabling storage consolidation, which reduces the number of systems under management and the robustness of disaster/recovery solutions.
InfiniVerse is Infinidat's petabyte-scale cloud-based analytics tool. It provides enterprise-wide predictive analytics, monitoring, and reporting on capacity and performance so that potential issues can be addressed before they become problems. InfiniVerse is part of our Elastic Data Fabric strategy and provides tight integration with cloud and transparent data mobility that together enable data to reside where it naturally belongs
Our InfiniBox interoperability matrix is currently 20 pages long and growing. The answer to most compatibility questions about operating systems, hypervisors, backup/restore, and databases is a resounding "yes". See the InfiniBox Interoperability Matrix for all the details.
We provide our customers with Technical Advisors. These TAs are experienced storage system engineers that provide our customers with named engineer support for the lifecycle of the installed system(s). This eliminates the customers' constant frustration of describing their environment to Support for the umpteenth time if they log a ticket. The TA provides a full spectrum of services as well as acting in the role of a customer advisor, which means they only install software updates that apply to their environment.
We have staffed and organized our support team to resolve problems quickly. Our level 1 support is staffed by highly trained engineers that use their product knowledge, rather than scripts, to quickly solve customer problems. And if they can not solve the problem, they escalate it directly to the development team with comprehensive troubleshooting notes. This further shrinks the time needed to resolve even difficult problems, which further improves customer satisfaction by reducing the time needed to resolve problems.
Interrupting Table 2 to mean that an InfiniBox will always consume less power than an AFA of equivalent capacity would be incorrect because InfiniBox needs lots of HDDs to deliver consistent sub-millisecond response times. Hence the InfiniBox environmental advantage only exists at scale.
Capacity and IOPS Costs
InfiniBox is a modern hybrid storage array that delivers sub-msec response times and PB-scale at lower $/IOPS and $/PB than our competitors. InfiniBox ease of use further reduces ownership costs along with our software update policies that keep older arrays architecturally current for longer periods of time, affordable maintenance, and disruptive economics that nearline HDDs make possible.
Abstracting storage decisions away from media technologies gives influencers and decision-makers (Storage Architects, Directors of Operations, and CIOs) the freedom to explore new storage architectures that deliver remarkably better capabilities and economics. It can also pressure incumbent vendors to provide users with lower prices by reducing the strength of their lock-ins. Users concerned about conversion and migration costs should know that Infinidat Professional Services are available to provide new customers with low-cost migrations. InfiniBox provides a disruptively better scale and financials versus AFAs. Against legacy hybrid arrays InfiniBox provides disruptively better ease-of-use and consistent sub-millisecond latencies without tuning.