{"id":39444,"date":"2019-10-31T22:32:47","date_gmt":"2019-10-31T19:32:47","guid":{"rendered":"https:\/\/prohoster.info\/blog\/programmno-opredelyaemye-shd-ili-chto-pogubilo-dinozavrov\/"},"modified":"2019-10-31T22:32:47","modified_gmt":"2019-10-31T19:32:47","slug":"programmno-opredelyaemye-shd-ili-chto-pogubilo-dinozavrov","status":"publish","type":"post","link":"https:\/\/prohoster.info\/en\/blog\/administrirovanie\/programmno-opredelyaemye-shd-ili-chto-pogubilo-dinozavrov","title":{"rendered":"Software-Defined Storage or What Killed the Dinosaurs?","gt_translate_keys":[{"key":"rendered","format":"text"}]},"content":{"rendered":"

\"Software-Defined
\n
\nOnce, they were at the top of the food chain. For millennia. Then the unimaginable happened: the sky was covered with clouds, and they ceased to exist. On the other side of the world, events unfolded that changed the climate: cloud cover increased. Dinosaurs became too large and slow: their attempts to survive were doomed to fail. Apex predators ruled the Earth for 100 million years, growing larger and becoming stronger. They evolved into what seemed like the perfect creature, sitting atop the food chain, but the Universe suddenly altered the appearance of our planet.
\n<\/a><\/noindex>
\nIronically, it was clouds that wiped dinosaurs off the face of the Earth 66 million years ago. Similarly, clouds today are destroying traditional storage systems that sit at the 'top of the food chain.' In both cases, the issue was not with the clouds themselves, but with the ability to adapt to a changing world. In the case of the dinosaurs, everything happened quickly: the devastating effects of the clouds set in within days or weeks of the meteor impact (or volcanic eruption\u2014the choice of theory is yours). For traditional storage systems, the process takes years, but it is certainly irreversible.<\/p>\n

Triassic Period: The Age of Great Iron and the Rise of Migrating Applications<\/h2>\n

\nSo what happened? In the existing ecosystem, there were entry-level and mid-range storage systems, enterprise-level systems, and direct-attached storage (DAS). These categories were defined by analysts, each having its own market volumes, cost, reliability, performance, and scalability metrics. Then something strange happened. <\/p>\n

The emergence of virtual machines meant that multiple applications, likely owned by different stakeholders, could run simultaneously on a single server. Such changes immediately called into question the future of Storage Area Networks (SANs) with direct connectivity. Then, the largest hyperscale infrastructure owners (hyperscalers) like Facebook, Google, eBay, and others, tired of paying exorbitant amounts for SANs, developed their own applications that ensured data availability on standard servers instead of large 'iron' SANs. Subsequently, Amazon introduced something strange to the market called Simple Storage Service (S3). It was neither a block nor a file, but something fundamentally new: one could no longer purchase a system, only acquire a service. Wait, what is that bright light visible in the sky? Another asteroid?<\/p>\n

The Jurassic Period: The Era of 'Good Enough Dinosaurs' <\/h2>\n

\nWe have entered a phase of SAN development with the ideology of 'good enough.' Customers using SANs, noticing what the hyperscalers achieved, began to question the fairness of the ten- or even hundredfold added value over hardware that they were paying for their corporate SANs. Mid-range arrays started to reclaim market share from high-end systems. Products like HPE 3PAR<\/a><\/noindex> demonstrated rapid growth. EMC Symmetrix, once the dominant array (from the word 'massive') in the enterprise class, still held some territory, but it was rapidly shrinking. Many users began migrating their data to AWS.<\/p>\n

On the other hand, SAN innovators started borrowing ideas from hyperscalers, using technologies from horizontally scalable distributed systems \u2013 an ideology opposed to vertical scaling. It is expected that the new SAN software will be able to run on standard servers, just like it does with hyperscalers. No more 10-100 times the price above the cost of the hardware itself. In theory, any servers can be used \u2013 the choice depends on your preferences. The era of Software-Defined Storage (SDS) has begun: clouds have closed the sky, temperatures have dropped, and the population of apex predators has started to decline.<\/p>\n

The Cretaceous Period: The Beginning of the Evolution of Software-Defined Storage <\/h2>\n

\nThe early days of software-defined storage (SDS) were tumultuous. There were many promises, but little was delivered. At the same time, a significant technological shift occurred: flash memory became a modern alternative to 'spinning rust' (HDD). This was a period marked by a plethora of SDS startups and easily accessible venture capital. Everything would have been great if it weren't for one problem: data storage requires serious commitment. It turned out that customers are quite attached to their data. If they lose access to it or find a couple of incorrect bits in terabytes of data, they become very anxious. Most startups did not survive. Clients received impressive functionalities, but the core tools often fell short. A bad recipe.<\/p>\n

Cenozoic Era: Storage arrays dominate<\/h2>\n

\nFew people discuss what happened afterward, as it\u2019s not very interesting \u2013 customers continue to purchase the same classic storage arrays. Certainly, those who moved their applications to the cloud also transitioned their data. But for the overwhelming majority of customers who do not want to go fully cloud or do not want to transition at all, Hewlett Packard Enterprise continued to offer classic arrays.<\/p>\n

We are living in 2019, so why does a multibillion-dollar storage business based on Y2K-era technology still exist? Because it works! Simply put, the demands of mission-critical applications were not met by products born from the hype wave. Products like HPE 3PAR remained the best options for enterprise clients, and the new evolution of the HPE 3PAR architecture\u2014 HPE Primera<\/a><\/noindex> \u2014only confirms this.<\/p>\n

In turn, the capabilities of software-defined storage were excellent: horizontal scalability, use of standard servers\u2026 But the price to pay for this was unstable availability, unpredictable performance, and specific scalability rules. <\/p>\n

The challenge of customer requirements is that they never get simpler. No one will say that data integrity loss or increased downtime is acceptable. This is why architecture is so important for storage systems, which must meet the demands of modern rapidly evolving data centers while preserving the key characteristics of enterprise-class storage.<\/p>\n

Tertiary period: the emergence of new life forms<\/h2>\n

\nLet\u2019s explore how one of the newcomers in the storage market \u2013 Datera \u2013 managed to deal with such a complex mix of historically established and new requirements for storage systems. Primarily by implementing an architecture focused on solving the dilemma described above. It is impossible to modify an old architecture to meet the tasks faced by modern data centers, just as it is impossible to adapt the architecture of a mid-range software-defined storage system to meet the requirements of enterprise-class systems: dinosaurs didn\u2019t become mammals because the temperature dropped. <\/p>\n

Building a solution that meets enterprise-level storage requirements while also considering the full value of the dynamism of modern data centers is a challenging task, but this is exactly what Datera aimed to achieve. Datera specialists worked on this for five years and found the recipe for 'cooking' enterprise-class software-defined storage.<\/p>\n

The main difficulty Datera faced was having to use the logical operator 'AND' instead of the noticeably simpler 'OR'. Stable availability, 'AND' predictable performance, 'AND' architectural scalability, 'AND' orchestration-as-code, 'AND' standardized hardware, 'AND' policy management implementation, 'AND' flexibility, 'AND' analytics-driven management, 'AND' security, 'AND' integration with open ecosystems. The logical operator 'AND' is one character longer than 'OR' \u2013 this is the key distinction.<\/p>\n

Quaternary period: modern data centers and the sharp climate change predetermine the development of software-defined storage.<\/h2>\n

\nSo how did Datera create an architecture that meets the requirements of traditional enterprise-class storage systems while also satisfying modern data center demands? It all comes back to that pesky operator 'AND'.<\/p>\n

There was no point in solving tasks one by one to meet individual requirements. The sum of these elements would not form a cohesive whole. As in any complex system, careful consideration of a balanced set of compromises was key. In their development, the experts at Datera focused on three core principles:<\/p>\n