Download the authoritative guide: Enterprise Data Storage 2018: Optimizing Your Storage Infrastructure
All-flash arrays (AFA) are popular for environments that need fast performance and low latency, such VDI and databases. But no all-flash array is automatically right for every environment. The trick to buying the right AFA for your needs is to a) know what your needs are and b) ask the AFA vendor the right questions so you’ll get accurate answers.
Before you make your AFA decision, ask your sales rep these twelve key questions - and make sure you get a straight answer!
Essential Questions to Ask When Selecting Flash Arrays
1. Capacity: What is your systems’ usable capacity figure? No, really.
Ascertain the usable capacity figure and understand what the vendor considers usable. Ask about dedupe and compression: When does dedupe occur? Is there a performance hit, and if so how much?https://o1.qnsr.com/log/p.gif?;n=203;c=204650394;s=9477;x=7936;f=201801171506010;u=j;z=TIMESTAMP;a=20392931;e=i
Also understand the vendor’s dedupe ratios for different workloads. All vendors figure in compression to dedupe ratios, figuring a 2:1 compression ratio on the duplicated data. So, if the vendor calculates a 10:1 compression on a database, they'll advertise it as 20:1. They’re not wrong, just be aware of the calculations that go into advertised ratios. Also ask them about different ratios for different types of workloads.
2. Performance: What are your IOPs and latency figures in my environment, not in your lab?
IOPs and latency are far less an issue in SSDs than HDDs, but they still matter. The ideal balance is high IOPs and low latency, but these numbers will not stay consistent across different types of workloads. Mixed, transactional and analytic workloads are going to take more IOPs and potentially more latency than well-behaved batch or sequential processing.
Get the real numbers for your AFAs real-world numbers for complex and high-performance workloads. If you add high-performance applications later, then expect to add more compute on your own dime. But if your vendor miscalculates performance figures after you accurately report your environment, then ask what they will do to remediate their mistake.
3. Throughput: How fast is data transfer really?
Throughput refers to data transfer speeds. (IOPs is the number of operations the storage system can perform per second, and latency is how quickly the data transfer starts.) AFAs value lies in high IOPs/low latency/fast throughput but the speeds you need in your environment may or may not be realistic with a given system. For example, different storage protocols can result in different performance results in AFAs. So can error code correction (ECC) overhead.
Block sizes are also an issue. Throughput speeds result from the number of I/O requests per second and the size of each input/output block. A 4K block yields fast throughput; a 256KB block slows it down. The difference may be negligible – or not. Understand how overhead and block sizes affect your AFAs throughput under different conditions.
4. Availability: How do you define “availability” and how do you prove yours?
99.999% is a common availability measurement for enterprise storage. Ask what tests your vendor is basing the percentage on. A knowledgeable sales engineer will pull out detailed tests to prove their claim.
Also ask how the storage architecture achieves this availability percentage: Software and hardware redundancy? HA features? Automated load balancing? Fault tolerant clusters? Always-on systems are designed to stay available during firmware upgrades, capacity and performance scaling and hot SSD replacements. Make sure yours does the same.
5. Durability: Do you insert advanced SSDs at the factory, or do you buy them at a garage sale?
I exaggerate about the garage sale. But some AFA vendors simply resell inexpensive third-party SSDs, going for the cheaper drives so they can undercut competitor prices. If you don’t ask the question, you might buy an AFA and assume that it includes garbage collection, packet striping, wear leveling and Error Correction Code (ECC).
Do not assume. If the vendor is going for low prices, they may be using cheaper SSDs without these critical processes. Ask your sales rep the type of SSDs they use and their native durability features.
6. Scalability: How efficiently do you scale, by which I mean I don’t want to buy a new system within a couple of years just to get more capacity.
Forklift storage upgrades used to be common. And feared. They are still a necessary evil when a business replaces one system with a different vendor system or replaces a long-outdated legacy system with a far newer one. What you do not want is to be stuck with the forklift upgrade because scalability is not what you expected.
Ask your vendor about their system’s scalability roadmap. If the system is scale-up, how much can you scale capacity without affecting performance? If the system is scale-out, how expensive is it to add nodes? How automated is the data movement process when you add or remove nodes?
7. Support: How much will support and maintenance cost after the honeymoon?
Maintenance flip occurs when storage vendors quote a low support price for the first one to two years, then significantly increase it in following years. Some customers live with it as a cost of doing business; others refresh early using higher support costs as a cost consideration. When researching your AFA, pin your vendor down on support and maintenance costs across the entire lifecycle. Also understand the costs of refreshing technology before the end of contract.
Ask about support time zones, expertise, and communication skills. Many end-users prefer to speak to support engineers who share their native language or are expert at a secondary language. Ask where the support is located, and if engineers can help or just read from a script. Also ask how much support is built into the AFA. Ask about native monitoring services that send alerts about performance impacts, failing SSDs and reaching capacity thresholds.
8. Integration: How well does the AFA integrate with applications?
It’s critical for AFAs to maintain high performance and low latency with hypervisors, management software, the cloud and data protection applications. Integrating with hypervisors seems to be a slam dunk, but the AFA needs to integrate not just with basic functions but advanced ones. For example, ask if and how the AFA supports advanced hypervisor functionality such as VMware VVOLs, or Hyper-V Storage Spaces Direct and Storage Replica. If the sales rep says that it does, ask how the AFA adds extra value. For example, does it add management functionality to Hyper-V with robust management toolsets, integrations, and monitoring tools?
And look for RESTful APIs for smooth integration with cloud-based automation platforms like OpenStack, HP Cloud Service Automation or Veeam Backup and Replication.
9. Management: What kinds of management tools do you offer?
Your AFA should integrate with existing management toolsets if you prefer to use them. You also have the right to expect high value native functionality on the all flash array. Once again, assume nothing. Make sure that the AFA comes with features like replication and snapshots, full-featured reporting, monitoring and alerts, autonomic self-healing, proven application and platform integration, global de-dupe and compression, caching and tiering optimized for SSDs, and thin technologies.
10. Simplicity: How easy is the array to manage?
AFAs will offer performance gains. But if IT spends more time managing the array than not, or if it’s all too easy to make mistakes on a complex management console, then you have not gained nearly as much advantage as you should. Look for automated management features like self-healing, policy-based management, linear performance and capacity scalability, hot SSD replacements and simplified firmware upgrades.
11. Connectivity: How connected is the array?
Multi-protocol conductivity and multiple OS support allow you to expand the AFA to its full capabilities. You might have no intention of using Fibre Channel if you’re an iSCSI shop, or iSCSI if you’re an FC SAN data center. Nevertheless, seriously consider looking for an AFA that comes with both Fibre Channel and iSCSI connections. This will allow you to protect your investment if down the road you decide to transition to one or the other. Make sure the network connections are high-speed. Also look at host operating system connections. Does the AFA support Windows, Linux and Unix equally well? It should.
12. Background: Is the vendor here for the long haul and did they originally design the AFA for flash-only?
Most well-known storage/AFA vendors are not likely to go under. But mergers and acquisitions can be difficult for end-users. Look at the world-shifting merger between Dell and EMC, and on a slightly smaller scale HPE’s acquisition of Nimble Storage and NetApp’s acquisition of flash maker SolidFire. Since a salesperson isn’t likely to confess her company is being bought up next month, do yourself a favor and keep up with M&A news.
Another consideration is buying an AFA from a new company. This isn’t necessarily a bad thing: every storage vendor started from scratch and the hottest new technologies can come from hot new teams. Do your homework: ask questions about the combined experience of the company’s founders, how involved they are in development, and if the company is an over-valued unicorn. (You might need to talk to a financial analyst for that.)
Also find out if the AFA is purpose-built for all-flash or repurposed from an HDD or hybrid array. Caching and tiering technologies that work well for disk do not necessarily work well for SSDs, which causes poor performance in your all flash array.