On January 21st, UPMEM’s team had the chance to participate to HiPEAC 2021 workshops on the theme of “Accelerating genomics/analytics application using UPMEM Processing In Memory”.
In this session we wanted to introduce 2 of the most advanced use cases using the UPMEM PIM technology while taking our time demonstrating ways to use the SDK for developers.
Use case #1: The acceleration of mapping & variant calling on PIM. for a whole human genome sequencing analysis (UPVC).
The generalisation of genomic analysis in medicine and biology relies on fast, accurate and cost effective computing architectures. UPMEM in cooperation with INRIA, proposed an acceleration based on PIM of the 2 fundamental steps for out of sequencers data analysis: mapping and variant calling.
UPVC takes advantage of the high PIM parallelism to complete the mapping stage. It consists in finding all potential positions of short reads on the map of the reference genome loaded in memory. The CPU relieved from the data heavy computation is available to perform advanced dynamic computing and complex tasks such as alignment with variant calling.
Use case #2: Index Search engine with PIM for a document database
An index search engine is a program that searches for and identifies items in a database that correspond to keywords or characters specified by the user, used especially for finding information, documents or web pages. Commercial search engines are estimated to process billions of queries daily and present a challenge in both capacity and performance to constantly provide a seemingless experience to its users.
This implementation work was realised under the guidance of a worldwide search engine leader to improve performances for a chain of 5 words query in an indexed document database. Aside from the specific algorithmic strategy, this work reveals the benefits in terms of latency and throughput when running a database search over a PIM architecture.
This workshop is organised in a way that attendants can understand how to make the most out of this novel architecture and apply those general principles to their own research.
Workshop summary:
- PIM Architecture overview
- SDK overview
- Code implementation with examples
- Introduction of the use cases
- Q&A