Myntra’s Hybrid Cloud implementation makes infra capable of 20x growth

by May 7, 2015 0 comments

This deployment gives Myntra the ability to scale its systems within a day to several times of its average business day, without incurring a significant TCO.

The online fashion retailing industry in India is growing, and the fact can be attributed to the country’s large population on the Internet. The e-commerce market in India is the fastest in Asia Pacific region with CAGR (Compounded Annual Growth Rate) of over 57% between 2012-2016 – worth USD 2.5 billion in 2009, which went up to USD 6.3 billion in 2011, and to USD 14 billion in 2012 – and within the online retail industry, the online fashion segment is growing fast, according to Forrester Research.

Bangalore-headquartered Myntra.com is one of the leading fashion and lifestyle products e-tailing platform in India. Over the last few years, the business at Myntra has scaled rapidly due to the ever-increasing popularity of online shopping, increased Internet reach in the country and quality of service that Myntra been consistently providing to its customers. According to a recent study, India ranks second in consumers accessing the internet for online shopping through mobile devices, just after China.

Myntra was acquired by the India’s largest online marketplace player Flipkart last year to cater to growing segment of online shopping in India. If a report to be believed, the fashion portal is expecting to turn profitable by next year, which signifies a considerable growth of Myntra’s business. This rapid scale of business poses a significant challenge in correspondingly scaling the IT Infrastructure. This rapid scaling poses a significant challenge in scaling the IT Infrastructure proportionally to support the business.

The Challenges

The IT Infrastructure in a predictable way for a rapidly growing business was a key challenge for Myntra’s IT team. The seasonality of this business also leads to sudden surges in demand, which was further straining the IT infrastructure in unpredictable ways. The IT team of Myntra believed that rapidly scaling engineering infrastructure was not going to be easy if the entire setup continued to be hosted on physical servers only. This was typically due to the high lead-time in procuring additional physical machines, and exorbitant fixed and recurring Total Cost of Ownership (TCO) that comes with relying on physical infra. So, for scaling rapidly, one must depend on some kind of compute virtualization or cloud solution, which allows for augmenting capacity quickly, with predictable costs.

At the same time, for mission critical parts of the applications, like databases, load-balancers, firewalls, critical storage, etc, virtualization is known to present significant challenges in terms of predictability in performance and reliability. Such mission critical systems are best hosted on bare metal that allows for a more predictable and deterministic performance, and operations of the applications.

Given these two seemingly conflicting requirements, the IT team thought of an architecture that would allow the company to simultaneously use the physical Infra for reliability and performance requirements, and Virtual Infra for on-demand scaling. “Essentially, both forms of Infra must co-exist for us. In that light, this project was undertaken,” said Shamik Sharma, CTO, Myntra.

The Implementation

Myntra worked with its IT Infrastructure partners, Netmagic Solutions Pvt Ltd to implement the architecture. The core engineering infra of Myntra.com is hosted in the datacenters that are owned and managed by the Netmagic team. Until recently, Myntra only had physical servers hosted with Netmagic, and all infra expansion for scale needs could happen by adding more physical servers.

However, Netmagic has a compute solution available as a cloud offering. This offered Myntra an opportunity to look at the Netmagic Cloud to scale quickly for unexpected surges in demand, still keeping the mission critical systems in the physical infrastructure to allow for more reliability and predictability of core systems.

Working with the Netmagic team, Myntra created a seamless tunnel between its physical infrastructure and Netmagic’s Cloud Infrastructure, thereby achieving both the objectives of scale and reliability at a very reasonable TCO.

Myntra also integrated its own security systems on top of the existing security systems that Netmagic Cloud had. “This gave us an additional layer of security on the shared cloud infrastructure, thereby alleviating any concerns we had around the security of our infrastructure and information flowing through it in the Hybrid model,” said  Sharma.

The USP of this project is in marrying the two seemingly conflicting requirements of rapid scaling with reliability, predictability and performance. There are many Infrastructure-on-Cloud solutions available in the market, and they work very well to provide on-demand scale. But in order to build reliability, predictability and performance for the mission critical parts on such solutions, the application design becomes complex, and the TCO to have redundancy increases significantly beyond a scale. “Such ‘pure-cloud solutions work very well for early stage IT companies, but for slightly mature setups, a Hybird infrastructure gives a better ROI,” said Sharma.

Besides, a study released by IDG Connect on behalf of Oracle estimates that the cloud computing landscape of 2017 will increasingly be dominated by platform and database services supporting hybrid infrastructures. It also finds out that organisations are more likely to choose hybrid cloud when considering their next step in cloud computing. Deploying more hybrid cloud services (36 percent) was selected ahead of private (32 percent) and public cloud (17 percent) services.

While implementing, the team also faced a few key technical challenges  including ‘transparent to the application’,   for any scaling solution to work effectively, it is imperative that the application code is never aware of the underlying infrastructure. That means that the engineering applications running www.myntra.com should be completely unaware that they are co-hosted on a combination of physical and cloud infrastructure, otherwise every time the system has to be scaled, it will have a direct impact on the application itself, and the solution wouldn’t work.

The IT team at the company managed to achieve this by bringing the cloud infra in the same LAN segments as the physical infra. This allowed for communication between the two segments over LAN rather than the WAN or Internet, thereby making it transparent to the applications and other system software.

Also, while the cloud infra is expected to take care of the surges, it is not possible for an enterprise to continue to expand on the cloud forever, as the core systems that require reliability and predictability will also need infra augmentation, and that had to be done on the physical infra side. Thus, it was imperative that seamless cloud connectivity shouldn’t impose any limitation to the expansion requirements on the physical infra side. Myntra’s IT achieved by building a network fabric like architecture, which allows for seamless expansion of physical infra up to 100 times of the existing setup, without affecting physical or cloud infra or the applications.  

The Results

The company could reliably scale its infrastructure to cater to around 20 times growth in the business with just 1.5 times the cost, in less than two months time.

And going forward, this gives Myntra the ability to dynamically scale its systems within a day to several times of its average business day, without incurring a significant TCO. Also, this allows for efficiently scaling up specific systems within its huge application stack without getting blocked on procuring hardware. This opens up avenues for experimentation, building Proof-Of-Concepts, bringing more efficiency in the engineering and IT teams for development of applications.

 The key users of this system are the IT and application development and engineering teams. They are getting benefits through less lead time in procuring infra to either scale their systems for business demands or running experiments for early stage engineering ideas or proof of concepts.

No Comments so far

Jump into a conversation

No Comments Yet!

You can be the one to start a conversation.

<