Search

AI Infrastructure plays a key role in the speed and cost-competitiveness of developing and deploying advanced AI models. The current demand for powerful AI infrastructure for model training is driven by the emergence of generative AI and foundational models, where on occasion thousands of GPUs must cooperate on a single training job for the model to be trained in a reasonable time. Delivering efficient and high-performing AI training requires an end-to-end solution that combines hardware, software and holistic telemetry to cater for multiple types of AI workloads. In this report, we describe IBM's hybrid cloud infrastructure that powers our generative AI model development. This infrastructure includes (1) Vela: an AI-optimized supercomputing capability directly integrated into the IBM Cloud, delivering scalable, dynamic, multi-tenant and geographically distributed infrastructure for large-scale model training and other AI workflow steps and (2) Blue Vela: a large-scale, purpose-built, on-premises hosting environment that is optimized to support our largest and most ambitious AI model training tasks. Vela provides IBM with the dual benefit of high performance for internal use along with the flexibility to adapt to an evolving commercial landscape. Blue Vela provides us with the benefits of rapid development of our largest and most ambitious models, as well as future-proofing against the evolving model landscape in the industry. Taken together, they provide IBM with the ability to rapidly innovate in the development of both AI models and commercial offerings., Comment: Corresponding Authors: Talia Gershon, Seetharami Seelam,Brian Belgodere, Milton Bonilla

Extreme dynamic heterogeneity in high performance computing systems and the convergence of traditional HPC with new simulation, analysis, and data science approaches impose increasingly more complex requirements on resource and job management software (RJMS). However, there is a paucity of RJMS techniques that can solve key technical challenges associated with those new requirements, particularly when they are coupled. In this paper, we propose a novel dynamic and multi-level resource model approach to address three key well-known challenges individually and in combination: i.e., 1) RJMS dynamism to facilitate job and workflow adaptability, 2) integration of specialized external resources (e.g. user-centric cloud bursting), and 3) scheduling cloud orchestration framework tasks. The core idea is to combine a dynamic directed graph resource model with fully hierarchical scheduling to provide a unified solution to all three key challenges. Our empirical and analytical evaluations of the solution using our prototype extension to Fluxion, a production hierarchical graph-based scheduler, suggest that our unified solution can significantly improve flexibility, performance and scalability across all three problems in comparison to limited traditional approaches., Comment: 11 pages, four figures, five tables

In this report, we present a new programming model based on Pipelines and Operators, which are the building blocks of programs written in PiCo, a DSL for Data Analytics Pipelines. In the model we propose, we use the term Pipeline to denote a workflow that processes data collections -- rather than a computational process -- as is common in the data processing community. The novelty with respect to other frameworks is that all PiCo operators are polymorphic with respect to data types. This makes it possible to 1) re-use the same algorithms and pipelines on different data models (e.g., streams, lists, sets, etc); 2) reuse the same operators in different contexts, and 3) update operators without affecting the calling context, i.e., the previous and following stages in the pipeline. Notice that in other mainstream frameworks, such as Spark, the update of a pipeline by changing a transformation with another is not necessarily trivial, since it may require the development of an input and output proxy to adapt the new transformation for the calling context. In the same line, we provide a formal framework (i.e., typing and semantics) that characterizes programs from the perspective of how they transform the data structures they process -- rather than the computational processes they represent. This approach allows to reason about programs at an abstract level, without taking into account any aspect from the underlying execution model or implementation., Comment: 24 pages

In the world of Big Data analytics, there is a series of tools aiming at simplifying programming applications to be executed on clusters. Although each tool claims to provide better programming, data and execution models, for which only informal (and often confusing) semantics is generally provided, all share a common underlying model, namely, the Dataflow model. The Dataflow model we propose shows how various tools share the same expressiveness at different levels of abstraction. The contribution of this work is twofold: first, we show that the proposed model is (at least) as general as existing batch and streaming frameworks (e.g., Spark, Flink, Storm), thus making it easier to understand high-level data-processing applications written in such frameworks. Second, we provide a layered model that can represent tools and applications following the Dataflow paradigm and we show how the analyzed tools fit in each level., Comment: 19 pages, 6 figures, 2 tables, In Proc. of the 9th Intl Symposium on High-Level Parallel Programming and Applications (HLPP), July 4-5 2016, Muenster, Germany