What should be an ideal storage system in 5 years?
What should be an ideal storage system? Sergey Yudin, Head of Data Storage Systems Management Group of Centralized Solutions Development Department of X5 Group, shared his opinion and experience: "Storage systems can be different. They differ in the type of storage or media, connection interface or access protocol. They are used both by ordinary people, for example, to store archives of photos, and by businesses of all kinds".
For the private user and small business, the cost of the solution, the ease of setup and installation, the ability to quickly replace components, for example, to buy at the nearest store are especially important. SOHO or Open Source solutions and cloud services are most suitable for it. Speed, broad functionality, compatibility and low downtime costs are important for medium-sized companies. These can be midrange storage, hyperconverged solutions (SDS) or Open Source solutions.
Large companies have the highest requirements: they are interested in reliability and TCO. Their choices are High-end solutions, clusters, distributed systems, S3 object systems. "In such companies, it is extremely difficult to implement super-modern functionality. Communication is seriously lengthened, shared access and rights tincture become more complicated," says Sergey Yudin. Thus, there is no ideal data storage system - each system must solve specific problems, the expert summarizes.
Roman Varnikov, Technical Director of Faberlic, used the example of his company to illustrate Sergey Yudin's presentation. He told that the company's main sales channel is the site. Its infrastructure has been formed for a long time, so it has a variety of solutions. This also applies to data storage systems. Against the backdrop of business growth, the question arose of modernizing storage systems for the internal cloud. The main requirements: conversion of CAPEX to OPEX, low total cost of storage, support for Faberlic Cloud, team competence and responsibility for business continuity.
As a result, an HPE solution was chosen, consisting of: 3PAR 8200 multisystem 3.3.1.648 (MU5), Brocade SAN fabric v8.2.2b Power Pack, FT over SAN fabric 12km + 25km, Remote Copy/Replication Groups, Priority Optimization. VMware, HyperV, Windows, Linux software was used.
Faberlic plans to further develop solutions based on the solutions of vendors with a strong position in the Russian Federation, to expand partnerships with Cloud providers, to focus on equipment with a long life cycle.
Sergey Platonov, RAIDIX development director, told about what awaits the storage market in the next five years. The list of today's challenges includes growing volumes of data and the speed of their processing, data collection at the edge of the environment (Edge) and their processing for making RealTime decisions, Cloud Native approach to application development and Multi-cloud development.
State of the storage market today
The market needs new solutions, said Sergey Platonov. For example, in data centers that use hyperconvergent approach, the problem of Dark Flash occurs, when users cannot predict what resources they will need in the near future. Because of this, they have to reserve them with a reserve of 30% to 70% according to various vendors' estimates. The way out is resource disaggregation, where instead of hyperconverged structures, separate modules are created that are delivered to the user using special software via high-performance networks. There are many startups in this market that offer different solutions. However, not all of them are efficient enough.
Specialized processors for different tasks, new CXL interconnect, new memory hierarchy and shared memory between the machines, and a data storage system that can be given a part of the tasks to be solved by Computational storage will help to solve the problem.
According to Sergey Platonov, the storage architecture of the future must include software based on the Intel SPDK and PMDK approach, which will run in containerized environments on DPU/IPU and BareMetal. It will also use non-volatile memory for Metadata, Cache, write optimization on QLCNAND,ZNS, SMR. It should also integrate with the gas pedal pipeline - functioning as distributed Computational Storage.