A New Opportunity for Smartphones
When you’ve finished with your smartphone, it typically finds its way into a drawer, joins the expanding second-hand market, or possibly ends up in a recycling bin. Nevertheless, it acts like a computer and, when paired with similar devices, can provide significant processing capabilities. Researchers at the University of California, San Diego, in partnership with Google, are set to launch an unconventional compute cluster not built from standard servers, but utilizing 2,000 retired phones. Their aim is to illustrate how these gadgets can continue functioning as an affordable, environmentally-friendly computing solution after their initial users have discarded them for a new device to scroll through TikTok.
The Visionaries Behind the Concept
“This initiative originated from Jennifer Switzer, a past PhD candidate at UCSD who is currently pursuing a post-doctoral position at Google,” Ryan Kastner, an associate professor of computer science at UCSD, informed GadgetLad. Specifically, UCSD will utilize 2,000 Pixel Fold smartphones provided by Google.
The Cost of Upgrade Culture
Google projects that most individuals replace their phones approximately every four years. Even though the physical device and battery may exhibit some signs of aging from their years of operation, their essential computing capabilities stay intact. “There’s just a huge amount of discarded compute, and recycling is not a viable solution for many of these smartphones,” Kastner remarked, further noting that Switzer began by assembling a few small clusters with smartphones to validate the idea. Since then, the project’s scope has expanded significantly.
Challenges in Repurposing Devices
According to the Chocolate Factory, the motherboard accounts for roughly 50 percent of the smartphone’s embodied carbon. A significant portion of initial tests utilized unaltered smartphones, Kastner remarked, but the team quickly realized this was neither practical nor safe. “In some early discussions with Google, their engineers mentioned that if you plan to install these in the data center, those batteries are a no-go—there are many concerns—because they pose fire risks,” he stated. Some of this work was conducted by researchers, including Switzer and fellow UCSD computer science professor Patrick Pannuto, but for the complete rollout this fall, Kastner indicated that Google is collaborating with a third party to remove the motherboards from their casings.
Powerful Processors in a Fresh Framework
Once the motherboard of the phones has been removed from their housings, the researchers believe that the chips inside are still more than capable enough for various tasks. In many situations, the single-threaded performance of these chips rivals, if not surpasses, that of many-core data center chips. The Pixel Fold smartphones, which will constitute the cluster’s foundation, are powered by a Google Tensor G2 processor featuring two 2.85 GHz Cortex-X1, two 2.35 GHz Cortex-A78, and four 1.80 GHz Cortex-A55 Arm cores, a Mali-G710 MP7 GPU, and 12 GB of RAM.
Assessing the Old Devices
Initial benchmarking using the SPEC suite indicates that 25-50 phones should provide performance comparable to that of a traditional server. The primary challenge instead lies in distributing workloads across numerous devices, each equipped with a few cores of different types, and most offering 8-12 GB of memory. UCSD researchers are tackling this issue from various angles. The first approach focuses on targeting applications that can comfortably fit within a single device. The second utilizes Kubernetes to manage container deployments across clusters of 25-50 phones.
Reviving Phones for New Purposes
For this initiative to succeed, the devices first need to have a Linux operating system installed that is appropriate for the task. While Android provides an excellent user experience on handheld devices, it is not designed for server use. In the blog post, researchers explain that Android contains features meant to prevent rogue applications from consuming excessive memory and draining battery life. In a server environment, these safety features become unnecessary. Kastner informed us that this was certainly not an easy feat, but the team has made consistent strides toward ensuring smooth operation of Linux on these devices, including compatibility with the phones’ built-in GPUs.
Addressing Connectivity and Security Issues
Access to certain functions, such as the integrated tensor processing unit in the chip, remains challenging. Connecting these devices will necessitate networking them together. Typically, these devices would communicate via cellular or Wi-Fi; however, at this scale, this is not only impractical but also raises security concerns, he clarified. Consequently, the team will implement PCBs that supply power while also facilitating wired Ethernet networking.
Handling Smaller Tasks with Mighty Phones
The researchers propose that many educational technology, grading, and research tasks commonly executed by universities in the cloud are sufficiently minor to operate on the cluster without any issues. “The vast majority of these applications can be handled by a single smartphone, with the standard grading backend functioning on small cloud instances,” a blog entry outlining the planned deployment indicates. “Initial experiments show that even a moderately-sized cluster of 20 phones can support peak submission rates for a class of 75 or more students.”
Pioneering New Avenues
“Many of the function-as-a-service workloads appear to make a lot of sense because they tend to be sporadic and don’t require extensive high-performance computing,” Kastner noted. In addition to conventional IT applications, the cluster will also facilitate exploration into parallel computing and systems programming, which sounds quite similar to the smartphone version of the Beowulf clusters from the ‘90s, where researchers scrambled together supercomputers from standard consumer PCs.
UCSD’s Aspirations
UCSD is also affiliated with the San Diego Supercomputing Center. Kastner mentioned that the plan is to make the cluster accessible to teams operating at the center, indicating that we might witness a High-Performance Linpack operation shortly. The complete smartphone cluster is anticipated to be launched this fall. Depending on the success of the initial phase, we’re informed that the cluster may expand even further.
Following Unique Paths
This is certainly not the only unconventional cluster we’ve encountered recently. Just north of San Diego, UC Santa Barbara set up what was then the largest Raspberry Pi cluster to date. This system, developed in cooperation with Oracle, comprised 1,050 Raspberry Pi 3B+ single board computers. More recently, a smaller cluster created by Gigabyte featured 40 Intel Lunar Lake notebook processors, each with eight cores and 32 GB of memory, all packed into a device the size of a pizza box.
Overview
Title: “From Pocket to Powerhouse: How Old Phones Get a New Lease on Life”
