Warehouse-Scale and Parallel Computing

Server racks with network delay

“Hiding” network latency for fast memory in data centers

A new system called Leap earned a Best Paper award at USENIX ATC ‘20 for producing remote memory access speed on par with local machines over data center networks.|Medium Read
Data warehouse

Enabling fairer data clusters for machine learning

Their findings reduce average job completion time by up to 95% when the system load is high, while treating every job fairly.|Medium Read

Undergraduate research on speeding up data centers earns ACM first prize

The student's project targets critical moments where the next instruction in a program is only available in a slower type of memory.|Short Read

Analytical model predicts exactly how much a piece of hardware will speed up data centers

The analytical model, called Accelerometer, can be applied in the early stages of an accelerator’s design to predict its effectiveness before ever being installed.|Medium Read

Putting hardware accelerators to work with automatic code translation

Most programs in use today have to be completely rewritten at a very low level to reap the benefits of hardware acceleration. This system demonstrates how to make that translation automatic.|Medium Read
Photo of Lingjia Tang

Creating more efficient data centers for AI

Tang’s project will redesign data center systems to support large-scale use of hardware accelerators to meet future computational demand.|Short Read
Prof. Mosharaf Chowdhury

Chowdhury receives VMWare Award to further research on cluster-wide memory efficiency

Chowdhury’s work has produced important results that can make memory in data centers both cheaper and more efficient. |Short Read
Large-scale data center

Chowdhury wins NSF CAREER award for making memory cheaper, more efficient in big data centers

Chowdhury connects all unused memory in a data cluster and treats it as a single unit.|Short Read
GPU cluster

Two solutions for GPU efficiency can boost AI performance

Chowdhury's lab multiplied the number of jobs a GPU cluster can finish in a set amount of time|Medium Read
Baris Kasikci

Speeding up code with clever data manipulation

Kasikci presents a method to improve a program’s ability to use data in a straightforward, efficient way|Short Read

Student earns Microsoft Fellowship for research in a new computing paradigm

Kassa is developing a framework that will look at the computations of an application and decide in real time which components will best handle it|Short Read
Large-scale data center

Designing a flexible future for massive data centers

A new approach recreates the power of a large server by linking up and pooling the resources of smaller computers with fast networking technology.|Medium Read