Why GHCL built a datacenter in a cyclone-prone, secluded sleepy little town—and saved crores.
Veraval, a small fishing port town on the coast of Gujarat, is a picture postcard of contradictions. It’s tucked between the soothing waters of the Arabian Sea and is home to the soot-filled chimneys of India’s biggest manufacturing plants. Its industries house the world’s most cutting-edge machinery, while its fishermen build dhows (fishing boats) with bare hands and measuring tapes.
This double-life makes Veraval interesting. But what’s significant about this port town is its geographical location. For one, it blesses the town’s surroundings with an abundance of limestone and salt resources—a prime ingredient in the manufacture of cement, and soda ash, and various other chemicals. Which is why, it attracts manufacturing giants like Aditya Birla Nuvo, Gujarat Ambuja Cement, and Gujarat Heavy Chemicals (GHCL).
GHCL, which manufactures soda ash—an essential ingredient for detergents, soaps, glass, and dyes—runs its entire chemical manufacturing business out of its Veraval plant, located barely a kilometer from the coast.
But Veraval’s picturesque setting hides a deadly reality: Its cyclone country. In 2005, a major cyclone hit the Veraval coast and severed GHCL’s communication lines, bringing business to a standstill for several hours. “The first three things that go down when a cyclone hits are power supply, communications, and network channels,” says Chandan Sinha, former CIO (who recently moved to Jindal Saw), GHCL.
Since the plant generates its own electricity, powering its operations during the cyclone wasn’t really an issue. Neither did the storm take down GHCL’s rudimentary ERP and standalone applications because they were run locally from the plant. Its link to the outside world, its VSAT, however, was a different story.
“We had a VSAT link which was working erratically but it couldn’t be used to access business applications. We could only use it for sporadic communication with other office locations during the cyclone,” recalls Sinha.
That incident made a deep impression on Sinha. At about the same time, GHCL was also planning to upgrade its ERP and wanted to build a full-fledged datacenter to provide the business with a strong back-end. The storm forced him to ask himself whether it was a great idea to set up the datacenter outside Veraval and connect users within the plan via a WAN.
The other option was building it in Veraval, in the middle of nowhere, far from any support. It was something to think about.
Point of Failure
The Rs 1,930-crore GHCL produces about 2,300 tons of finished goods a day. These goods require about 6,000 tons of raw material. If you add it up, that’s over 8,000 tons of material moving in and out of the plant—on an average—a day.
Much of that movement is monitored using IT. When trucks carrying raw materials enter the plant, weigh bridges weigh them and generate a logistic invoice. Similarly, trucks on the way out loaded with finished goods also need an invoice. With about 450 trucks entering and exiting the plant every day, it’s a lot of IT-dependant invoice creation.
During peak hours, Sinha says, systems generate one invoice a minute. So when systems go down, invoices pile up, bringing production to a halt. “Not to mention the long cue of trucks outside my factory gates, which opens up another set of problems,” says Sinha.
That made it imperative for GHCL to ensure that business users had access to information no matter what.
And that could only be achieved if Sinha built infrastructure in a way that users at the plant location could access applications over a LAN. This would guarantee that unless it was doomsday, his business would not come to a standstill.
But to achieve that Sinha would have to do something phenomenal. Like build a datacenter in a sleepy little town, prone to cyclones. For most CIOs, either one of those would have reason enough to put as much distance as they could between Veraval and their datacenters.
Going the Full Distance
Although GHCL’s plant in Veraval is well connected to important towns in Gujarat, its distance from the commercial centers of the state made it a risky location to set up a datacenter. The closest city to Veraval, Ahmedabad, is almost 400 kilometers away, and the nearest metropolitan city, Mumbai, is 900 kilometers away. In case of an emergency, it could take anything between 12-24 hours for help to reach Veraval.
“Getting hardware to Veraval was not a challenge, but we had to factor in the possibility of a server breaking down. Then it would take us a minimum of 12 hours to source it from the nearest city,” says Sinha.
The biggest task for Sinha was to choose hardware that was resilient and could withstand the vagaries of a sea-side location including high humidity levels and salty air.
Not only did the servers need to be weather-tested but Veraval’s distance from other cities also made it necessary to ensure that the servers had the feature of remote troubleshooting.
“In case of an emergency, someone should be able to perform the first level of trouble shooting without having to be physically present at the datacenter,” Sinha says.
He says that the exercise of choosing servers was extremely challenging because while he could find servers that could withstand extreme climate, servers that allowed remote assistance weren’t readily available when he was conceptualizing the project.
Sinha says he spent about six months evaluating various products. He spoke with his peers from similar verticals operating in, more or less, similar conditions to understand their challenges. “From technical forums to long conversations with solution providers, I checked the performance of the servers and other hardware components,” says Sinha.
He tried six ways from Sunday to create a highly resilient datacenter. But he also knew that he needed to chalk out a response plan with the smallest possible downtime in case something within the datacenter went askew.
Which is why, Sinha created a mirror—a backup—for every server in the datacenter. “This n+1 provisioning ensured that even if a server broke down, its mirror would be up and running in the least possible time,” he says. And if that went down, Sinha could plug in to his DR site in Ahmedabad.
A rather expensive arrangement, Sinha managed to offset the price of over provisioning by leveraging a few advantages he had at the plant location.
Utilizing the plant at Veraval to build his datacenter meant that GHCL had enough land to spare for a 2,000 sq. ft. datacenter without having to dole out a dime for it. The real estate benefit itself brought down the cost of the project significantly. That the plant has its own captive power generation capability meant uninterrupted power supply at no additional cost.
“Real estate is a considerable expense when setting up a datacenter. Having a plant location at Veraval meant that we wouldn’t have to incur any extra cost on real estate for the datacenter,” says Sinha.
And despite the over-provisioning, the datacenter cost GHCL only about Rs 5 crore. “A datacenter in Noida or Gurgaon would have cost about Rs 8 crore to build. Not just that, my expenditure on power, cooling, and real estate would have led to at least three times more opex than what I currently spend,” says Sinha.
And it also wouldn’t have provided Sinha with almost free air-conditioning. The plant’s geographical location in Veraval—which enjoys a cool sea breeze—helps Sinha leverage the concept of ambient cooling for the datacenter. This meant he could reduce the number of ACs.
With the infrastructure in place, Sinha had only one more hurdle to jump: Getting talent to man his datacenter in the boondocks. It was a pretty tall hurdle.
Fishing for Talent
In retrospect, getting the infrastructure bits right was a relatively easier problem to tackle, compared to finding people willing to work in a desolate location, far away from their families, says Sinha.
Despite being an industrial hub, Veraval has a population of only 1.5 lakh and is bereft of the glitz that metros and tier 1 cities offer. So even if he did manage to find people and convince them to move to Veraval, he knew he would have to be ready for a high attrition rate.
“I needed a team that was dedicated; that’s what it takes to work under harsh conditions, in a small, little industrial town in the middle of nowhere. Attrition could become a serious problem because replacing a resource would be even more difficult,” he says.
While talking to his team members and peers in Gujarat, Sinha unearthed a trait of the local population that he realized he could use to his benefit.
“I found out that people from this region are extremely family-oriented and like to live close to their homes. Also, hiring people locally would ensure I picked people who were well-adapted to Veravel’s extreme weather conditions,” he says.
Both Ahmedabad and Baroda have a lot of IT and engineering colleges. Sinha decided to go on a campus hunt and handpicked qualified candidates to build a 15 man-strong team at Veraval. He says the selection process tilted in favor of candidates from the Junagarh side of the state—Veraval falls in the Junagarh district. This would minimize the risk of attrition.
Sinha had to invest considerable time and effort in ensuring that the recruits, fresh out of college, were trained adequately and could handle complicated systems like ERP. Today, it’s evident that Sinha’s bet paid off. He says his team hasn’t seen even a 1 percent attrition rate since the project took off.
“The path I chose was the least obvious one, but saying no to the ordinary is what has made all the difference,” he says.
That, it has.
United Breweries faced a challenge: Tracking and monitoring company assets deployed to supply draught beer to the bars. RFID technology puts all problems to rest.
Facebook at Work, the company's social network for business, has been available as a limited pilot for about a year, but the support of Royal Bank of Scotland helps legitimize the soon-to-be-released product for security-conscious organizations.
To make drilling more efficient during a commodities industry downturn, Laredo Petroleum is turning to cloud applications from a Silicon Valley startup to correlate production and operational costs.