NEW YORK — If you’re an IT manager, cloud computing will fundamentally change your job, said Hal Stern, Sun vice president of engineering, in a speech at the technology management conference of the Securities Industry and Financial Markets Association (SIFMA) here Wednesday.
“With services, we are leaving the hardware world,” Stern said.
He meant that system administrators will monitor the network and not its hardware components. “We still need sysadmins but we don’t need them running around the datacenter with a socket wrench,” he said. “Instead, they will use telemetry and tools to assess capacity, security, and performance.”
But what is the cloud? According to Gartner, the cloud has five attributes. It is service-based. It is scalable and elastic, able to add and remove infrastructure as needed. It uses shared infrastructure to build economies of scale. It is metered and users pay according to usage. Most importantly, of course, it uses Internet technologies.
Some companies don’t want to share the infrastructure, so they build what is called private clouds. Others focus on price, and are willing to share the cloud infrastructure with other companies in cheaper public clouds.
“In Sun’s view, there will be many clouds. There will be private clouds and public clouds and a spectrum of clouds in between them, even though at the moment the distribution of clouds is barbell shaped,” Stern said.
Tension between developers and deployers
For IT managers in general and system administrators in particular, cloud computing can solve one nagging headache, Stern said. “Cloud computing can help solve the tension between developers and deployers. This tension has existed since the Garden of Eden. Why would the apple be there if not for developers to play with it,” he joked, inferring that IT administrators are God.
He said, “IT administrators ask: why are users so needy? How can I audit what they’re doing? Why do they need so many versions?”
All of this is easier in the cloud. Applications can be monitored and deployed better, depending on a company’s needs.
“Startups see the cloud as a way to spend money on salaries, developers, and beer — and not on infrastructure,” Stern said.
The cloud makes it easier to monitor usage, but paying according to use isn’t always cheaper, Stern warned. He noted that if you drive a car every day, you should buy one, but if you drive a car occasionally, it should make financial sense to rent one.
Sometimes, business managers use the cloud to avoid IT. Stern pointed to the story of New York Times data architect Derek Gottfrid who used Amazon’s AWS and the Hadoop parallel data processing architecture to turn 70 years of newspapers into the TimesMachine archive.
“Eyebrows were raised when Gottfrid did an end run around the IT department,” Stern said.”
Next page: The structure of a cloud
Page 2 of 2
The structure of a cloud
Stern said that IT will need to change how it thinks of assets in order to manage the coming multi-cloud infrastructure. Building on a recent blog post and speech, Stern said that infrastructure is now addressed through URLs and APIs and not through read and write commands.
He said it takes a surprising weight of manuals to learn how to use non-cloud infrastructure to create a Hello world program, a simple piece of software that prints the words “Hello, world”
to a printer or display. “As complexity increases, interest declines.”
In contrast, cloud applications can be built quickly. In response, sysadmins will need to abandon their favorite equation for measuring reliability.
“In the nineties, we measured reliability as: MTBF/(MTBF + MTTR) where MTBF is mean time between failure and MTTR is mean time to restart,” Stern said.
He said that the equation usually yielded a number with a lot of nines in it. He said that as administrators looked at the equation, they worried about what if all the downtime came at once, or about how the equation would be effected if a server went down and stayed down for 50 minutes even just once during the year.
“So we bought hardware to lower the mean time to restart. We bought RAID and SAN and clusters,” he said. “But software and deployment are also affecting reliability.”
Sysadmins need to measure service performance by tracking a series of KPIs Stern called PIPE. “Sysadmins need to measure predictability, integrity, productivity, and efficiency,” he said.
He added that sysadmins will measure the efficiency of the datacenter in terms of throughput per dollar and work per watt.
Sysadmins will not track hardware performance in the cloud. “We do not see the underlying hardware even in the private cloud and we may not know what the hardware is in the public cloud.”
Sysadmins no longer worry about replacing hardware. “If you have 10,000 servers, something’s always failing” said Stern. “But there’s also often a software failure.”
He said that instead of focusing on mean time to reboot a server, sysadmins will focus on time taken to recognize a problem. They will be purging old instances of an application and will be adding more instances if they need to.
He said that in the cloud, a poor response time is virtually the same as failure.
He added that although this change in job function won’t be easy, there’s a massive ecosystem of competing cloud providers who are eager to help.
“Remember: think ‘recognize and restart’ and not nines in terms of reliability,” he concluded.