Huawei CEO Ken Hu isn’t just promising to deliver a technology that’s nowhere defined. He’s reducing network innovation down to a question of mere speed. The mobile industry should have much bigger priorities.
Yes, 4G and LTE networks are huge improvements over previous-generation mobile networks, but they’re not about to cure mobile-data woes without some smart development to help them. Speaking at GigaOM Mobilize on Thursday, Aspera’s Serban Simu and Akamai’s (s akam) Kris Alexander espoused the value of both intelligent network design and intelligent app designers to begin with.
The problem with transferring large files such as movies over mobile networks really boils down to distance and congestion. As Alexander explained, even a 300-kilobit web site can take a long time to load because it might involve 30 different requests that each must make the roundtrip from phone to mobile network to the internet and back. Right now, he said, web pages for major companies that take 3 seconds to load over the web take 9 seconds to load on a mobile device, which is a huge problem if companies don’t want users abandoning their sites.
Aspera’s Simu pointed to congestion as a major inhibitor of rich mobile experiences. Although 4G networks can boost performance of his company’s file-transfer tool by three to four times over 3G networks, Simu said, it’s a volatile medium on which performance is inconsistent, in large part because a network can become congested in just fractions of a second.
One part of the solution will come from carriers, device manufacturers, and middlemen such as Aspera and Akamai that can make data transfer that much more efficient. Referencing those 30 roundtrips to load a standard web site, Alexander asked, “What if I could do it in two or three? What if i could do it in one?”
Or if devices and servers were both speaking the same language, he added, perhaps they could intelligently decide to take shortcuts that deliver content in a manner that accounts for the chatty nature of TCP/IP networks. Akamai, he noted, it working on all sorts of tactics for improving the performance of mobile data networks, and it plans to make them available industry-wide.
Simu suggested the idea of caching data on devices themselves, noting that a standard device can easily hold enough properly encoded HD video to let someone stream episodes of a television series and keep only the data that’s needed to watch what’s coming next.
But the real trick is for app developers to get smart about programming. They can’t rely on bandwidth to improve, and even tools like what Aspera and Akamai are doing can only improve certain aspects of a mobile app’s performance. A beautiful UI is great, Simu said, but maybe not as important as reducing those 50 REST calls down to 1, or figuring out whether the amount of data an app has to send in order to be useful is even feasible at all.
Former Facebook engineers Eric Frenkiel and Nikita Shamgunov launched a startup called MemSQL that seeks to speed relational databases by taking a page out of the Facebook playbook. MemSQL boosts performance by keeping data in memory and converting SQL code into faster C++.
Amazon DynamoDB brings the power of the cloud to NoSQL databases, but the storage looks expensive for big data apps and most developers are still unclear about what “eventual consistency” means.
According to Lou Modano, head of global infrastructure for NASDAQ OMX, there is “a lot of opportunity for new [hardware] players” to feed a ceaseless demand for ever lower latency in electronic trading systems. When you rely entirely on electronic trading, a faster system is key.
WAN-optimization startup Infineta has raised $15 million for its product that speeds traffic flows between data centers. Unlike many WAN-optimization products that speed traffic between user sites and a data center, Infineta targets data moving between data centers at up to 10 GbE speeds.
Uploading video to the web can be a painfully long process even with a high-speed connection. I spoke with Shane Russell of Microsoft’s VidLab, who shared with me how his team delivers content to the Zune Marketplace and the Xbox LIVE service while slashing network latency.
Hardware rarely comes up in discussions about big data, save for those centered on data warehouse appliances. But the omission hardly means hardware is irrelevant. In fact, big gear might become a big deal as companies look to bolster the performance of their big data systems.
There are two good features today about the increasingly hot colocation business. What’s interesting to me is that companies appear to be addressing their ramped-up computing needs with externally housed physical servers rather than pure virtual servers (aka cloud computing). Although some colocation space is taken up by cloud platforms, it would be great to know what that percentage is. That answer could prove whether companies are buying into cloud computing, or just the not-managing-a-data-center part of it. As for high-frequency trading, well, that might never go virtual, and data centers proximately located next to exchanges likely will remain full for the foreseeable future.