A paper titled 'MFence: Defending Against Memory Access Interference in a Disaggregated Cloud Memory Platform', written by graduate students Lee Jin-hoon (master's, 4th semester; supervisor: Professor Kim Young-jae), Jung Yeon-woo (doctoral, 6th semester; supervisor: Professor Park Sung-yong) and Safdar Jamil (combined master's–doctoral, 10th semester; supervisor: Professor Kim Young-jae) of the Department of Computer Science & Engineering under the supervision of Professor Kim Young-jae (corresponding author), has been accepted for publication at the ACM/SIGAPP Symposium on Applied Computing, Operating Systems Track, listed at an adjusted IF of 1 among the outstanding international conferences in computer science under the BK21 Plus programme. The paper proposes MFence, a cache partitioning mechanism that resolves the problem of unfair memory occupancy of shared cache between processes in VM-based disaggregated cloud memory platform environments. MFence is a kernel module supporting independently partitioned cache regions that completely eliminates cache interference between processes.

A VM-based disaggregated cloud memory platform is an architecture in which two or more nodes can borrow or lend one another's memory, improving application performance by increasing memory utilization and capacity. Because this architecture uses local memory on a given node as a cache and remote nodes as swap space (OS swap), cache contention arises within local memory. The paper proposes MFence, an architecture that prevents cache contention using two mechanisms, hypercall and gCR3.

MFence was shown to enable fair cache partitioning with low overhead and to bring performance gains. As a representative example, when aggregation and grep workloads ran simultaneously, cache partitioning through MFence yielded a 14.4% performance improvement.