System Bits: Sept. 5

Reducing power consumption of datacenter caches
As is commonly understood, most websites store data in databases, and since database queries are relatively slow, most sites also maintain so-called cache servers, which list the results of common queries for faster access, researchers from MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) reminded. They noted that a data center for a major web service such as Google or Facebook might have as many as 1,000 servers dedicated just to caching, and the cache servers generally use random-access memory (RAM), which is fast but expensive and power-hungry. To this point, the MIT team has devised a system for data center caching that instead uses flash memory, the kind of memory used in most smartphones.

The team said that per gigabyte of memory, flash consumes about 5 percent as much energy as RAM and costs about one-tenth as much. It also has about 100 times the storage density, meaning that more data can be crammed into a smaller space. In addition to costing less and consuming less power, a flash caching system could dramatically reduce the number of cache servers required by a data center.

Researchers from CSAIL have devised a new system for data center caching that uses flash memory. In addition to costing less and consuming less power, a flash caching system could dramatically reduce the number of cache servers required by a data center. 
(Source: MIT)

The drawback to flash is that it’s much slower than RAM, they said, and this is where the disbelief comes in but slow as it is relative to DRAM, flash access is still much faster than human reactions to new sensory stimuli, the researchers asserted. They believe users won’t notice the difference between a request that takes .0002 seconds to process — a typical round-trip travel time over the internet — and one that takes .0004 seconds because it involves a flash query.

More importantly, the system needs to keep up with the requests flooding the data center. The CSAIL researchers’ system, BlueCache, does that by using pipelining: Before a flash-based cache server returns the result of the first query to reach it, it can begin executing the next 10,000 queries. The first query might take 200 microseconds to process, but the responses to the succeeding ones will emerge at .02-microsecond intervals, they explained.

At the same time, the CSAIL researchers had to deploy some clever engineering tricks to make flash caching competitive with DRAM caching. In tests, they compared BlueCache to what might be called the default implementation of a flash-based cache server, which is simply a data-center database server configured for caching. They reported that BlueCache was 4.2 times as fast as the default implementation.

App could use smartphone selfies for pancreatic cancer screening
It’s no secret that pancreatic cancer has a dismal prognoses — a five-year survival rate of 9 percent — due in part to the fact that there are no telltale symptoms or non-invasive screening tools to catch a tumor before it spreads. University of Washington researchers have set out to change that and are developing an app that could allow people to easily screen for pancreatic cancer and other diseases — by snapping a smartphone selfie.

BiliScreen is a new smartphone app that is designed to screen for pancreatic cancer by having users snap a selfie. It’s shown here with a 3-D printed box that helps control lighting conditions to detect signs of jaundice in a person’s eye.
(Source: University of Washington)

‘BiliScreen‘ uses a smartphone camera, computer vision algorithms and machine learning tools to detect increased bilirubin levels in a person’s sclera, or the white part of the eye given that one of the earliest symptoms of pancreatic cancer, as well as other diseases, is jaundice, a yellow discoloration of the skin and eyes caused by a buildup of bilirubin in the blood. The ability to detect signs of jaundice when bilirubin levels are minimally elevated — but before they’re visible to the naked eye — could enable an entirely new screening program for at-risk individuals.

In an initial clinical study of 70 people, the BiliScreen app — used in conjunction with a 3D printed box that controls the eye’s exposure to light — correctly identified cases of concern 89.7 percent of the time, compared to the blood test currently used.

“The problem with pancreatic cancer is that by the time you’re symptomatic, it’s frequently too late,” said Alex Mariakakis, a doctoral student at the Paul G. Allen School of Computer Science & Engineering. “The hope is that if people can do this simple test once a month — in the privacy of their own homes — some might catch the disease early enough to undergo treatment that could save their lives.”

BiliScreen builds on earlier work from the UW’s Ubiquitous Computing Lab, which previously developed BiliCam, a smartphone app that screens for newborn jaundice by taking a picture of a baby’s skin. A recent study in the journal Pediatrics showed BiliCam provided accurate estimates of bilirubin levels in 530 infants.

BiliScreen uses a smartphone’s built-in camera and flash to collect pictures of a person’s eye as they snap a selfie. The team developed a computer vision system to automatically and effectively isolate the white parts of the eye, which is a valuable tool for medical diagnostics. The app then calculates the color information from the sclera — based on the wavelengths of light that are being reflected and absorbed — and correlates it with bilirubin levels using machine learning algorithms.

Doubling mobile web speed with software
Despite that most web traffic today comes from smartphones and tablets, the mobile web remains inconveniently slow. Even on fast 4G networks, a page takes 14 seconds to load on average—an eternity in today’s connected world. To combat this, a team of computer science researchers at the University of Michigan and MIT said they’ve developed a way to dramatically speed up the mobile web.

The ‘Vroom’ software prototype works by optimizing the end-to-end interaction between mobile devices and web servers. Tested on 100 popular news and sports websites, the researchers found that Vroom cut in half the median load time on landing pages—from 10 seconds to 5.

For the Fox News mobile site, (a) rendering of the above the fold content completes at 9.26s with VROOM; (b) with only HTTP/2 enabled, rendering is incomplete at that time and completes only later at 13.87s.
(Source: University of Michigan)

Harsha Madhyastha, U-M associate professor of computer science and engineering and one of Vroom’s developers said Vroom dramatically improves upon solutions such as proxy servers, which come with security and privacy concerns. “It complements solutions such as Google’s Accelerated Mobile Pages project, which requires web pages to be rewritten. For any particular version of a web page, Vroom optimizes the process of loading that page.”

A key reason for the lag on mobile sites is that, even when a user visits a mobile-optimized page, the browser must incrementally discover, download and process close to 100 URLs—the resources that constitute the page—before that page fully reveals itself. Vaspol Ruamviboonsuk, U-M doctoral student in computer science and engineering who led the development of Vroom explained, “A lot needs to be bound and assembled, especially on sports and news pages with live content and personalized ads. When a browser begins to load a page, all it knows is the main URL. Everything else, it has to discover on its own through multiple rounds of parsing and executing code to determine all the assets it needs.”

This back-and-forth is necessary because both the central processing units and the networks of mobile devices are much slower than their counterparts on desktop and laptop machines. As a result, the mobile device’s CPU sits idle and underutilized while requests and responses are transferred to servers over the cellular network.

One could rely on proxy servers to accelerate websites. Proxies essentially act as virtual CPUs, building out pages before transferring them to the browser. But they compromise security and privacy. They intercept HTTPS content and require access to a user’s cookies. In contrast, the new Vroom architecture bundles together resources that browsers will need to fully load pages. When a web server receives a request from a browser, in addition to returning the requested resource, the server also informs the browser about other dependent resources it will need to fetch.