Seminar Series Archive
Amirhossein Mirhosseini
University of Michigan, Ann Arbor
February 27, 2020
3:00pm - 4:00pm
Title:
Server Architectures for the Microservices Era
Abstract:
Modern Online Data Intensive (OLDI) cloud services (e.g., web search, social media) are built using distributed microservice architectures, where a complex application is decomposed into numerous discrete microservices that interact over high-performance data center networks using remote procedure calls (RPCs). Individual RPCs are often extremely short, ranging from single digits to hundreds of microseconds. Many cloud-based companies, including Amazon, LinkedIn, and Netflix have adopted microservice-based architectures.
In this seminar, I will talk about a few important challenges in the microservices era, and how we can address them in hardware/software. First, I will talk about the "Killer Microseconds" challenge, which refers to us-scale ``holes'' in CPU schedules caused by stalls to access fast I/O devices or brief idle times between requests in high throughput microservices. I will present our proposed heterogeneous server architecture, Duplexity, which employs aggressive multithreading to hide the latency of killer microseconds, without sacrificing the Quality-of-Service (QoS) of latency-sensitive microservices. Then, I will present Caper, which is an algorithmic scheduling framework to tackle the tail latency of microservices from a queuing perspective, and CaperCore, as a microarchitectural instantiation of the framework. Finally, I will conclude my talk by describing the ongoing and future directions towards designing highly efficient server architectures in the microservices era.
In this seminar, I will talk about a few important challenges in the microservices era, and how we can address them in hardware/software. First, I will talk about the "Killer Microseconds" challenge, which refers to us-scale ``holes'' in CPU schedules caused by stalls to access fast I/O devices or brief idle times between requests in high throughput microservices. I will present our proposed heterogeneous server architecture, Duplexity, which employs aggressive multithreading to hide the latency of killer microseconds, without sacrificing the Quality-of-Service (QoS) of latency-sensitive microservices. Then, I will present Caper, which is an algorithmic scheduling framework to tackle the tail latency of microservices from a queuing perspective, and CaperCore, as a microarchitectural instantiation of the framework. Finally, I will conclude my talk by describing the ongoing and future directions towards designing highly efficient server architectures in the microservices era.
Speaker Bio:
Amirhossein Mirhosseini is a PhD candidate of Computer Science and Engineering at the University of Michigan, Ann Arbor, and a visiting researcher at the Massachusetts Institute of Technology (MIT). His research centers on computer systems and architecture, with particular emphasis on designing hardware primitives for emerging software paradigms, including cloud microservices. He is the recipient of the best paper award at the International Symposium on Networks-on-Chip (NOCS), 2017.