The need for computing in the HEP community follows cycles of peaks and valleys mainly driven by conference dates, accelerator shutdown, holiday schedules, and other factors. Because of this, the ...classical method of provisioning these resources at providing facilities has drawbacks such as potential overprovisioning. As the appetite for computing increases, however, so does the need to maximize cost efficiency by developing a model for dynamically provisioning resources only when needed. To address this issue, the HEPCloud project was launched by the Fermilab Scientific Computing Division in June 2015. Its goal is to develop a facility that provides a common interface to a variety of resources, including local clusters, grids, high performance computers, and community and commercial Clouds. Initially targeted experiments include CMS and NOvA, as well as other Fermilab stakeholders. In its first phase, the project has demonstrated the use of the "elastic" provisioning model offered by commercial clouds, such as Amazon Web Services. In this model, resources are rented and provisioned automatically over the Internet upon request. In January 2016, the project demonstrated the ability to increase the total amount of global CMS resources by 58,000 cores from 150,000 cores - a 38 percent increase - in preparation for the Recontres de Moriond. In March 2016, the NOvA experiment has also demonstrated resource burst capabilities with an additional 7,300 cores, achieving a scale almost four times as large as the local allocated resources and utilizing the local AWS s3 storage to optimize data handling operations and costs. NOvA was using the same familiar services used for local computations, such as data handling and job submission, in preparation for the Neutrino 2016 conference. In both cases, the cost was contained by the use of the Amazon Spot Instance Market and the Decision Engine, a HEPCloud component that aims at minimizing cost and job interruption. This paper describes the Fermilab HEPCloud Facility and the challenges overcome for the CMS and NOvA communities.
How much higher can HTCondor fly? Fajardo, E M; Dost, J M; Holzman, B ...
Journal of physics. Conference series,
12/2015, Volume:
664, Issue:
6
Journal Article
Peer reviewed
Open access
The HTCondor high throughput computing system is heavily used in the high energy physics (HEP) community as the batch system for several Worldwide LHC Computing Grid (WLCG) resources. Moreover, it is ...the backbone of GlidelnWMS, the pilot system used by the computing organization of the Compact Muon Solenoid (CMS) experiment. To prepare for LHC Run 2, we probed the scalability limits of new versions and configurations of HTCondor with a goal of reaching 200,000 simultaneous running jobs in a single internationally distributed dynamic pool. In this paper, we first describe how we created an opportunistic distributed testbed capable of exercising runs with 200,000 simultaneous jobs without impacting production. This testbed methodology is appropriate not only for scale testing HTCondor, but potentially for many other services. In addition to the test conditions and the testbed topology, we include the suggested configuration options used to obtain the scaling results, and describe some of the changes to HTCondor inspired by our testing that enabled sustained operations at scales well beyond previous limits.
The Fermilab HEPCloud Facility Project has as its goal to extend the current Fermilab facility interface to provide transparent access to disparate resources including commercial and community ...clouds, grid federations, and HPC centers. This facility enables experiments to perform the full spectrum of computing tasks, including data-intensive simulation and reconstruction. We have evaluated the use of the commercial cloud to provide elasticity to respond to peaks of demand without overprovisioning local resources. Full scale data-intensive workflows have been successfully completed on Amazon Web Services for two High Energy Physics Experiments, CMS and NOνA, at the scale of 58000 simultaneous cores. This paper describes the significant improvements that were made to the virtual machine provisioning system, code caching system, and data movement system to accomplish this work. The virtual image provisioning and contextualization service was extended to multiple AWS regions, and to support experiment-specific data configurations. A prototype Decision Engine was written to determine the optimal availability zone and instance type to run on, minimizing cost and job interruptions. We have deployed a scalable on-demand caching service to deliver code and database information to jobs running on the commercial cloud. It uses the frontiersquid server and CERN VM File System (CVMFS) clients on EC2 instances and utilizes various services provided by AWS to build the infrastructure (stack). We discuss the architecture and load testing benchmarks on the squid servers. We also describe various approaches that were evaluated to transport experimental data to and from the cloud, and the optimal solutions that were used for the bulk of the data transport. Finally, we summarize lessons learned from this scale test, and our future plans to expand and improve the Fermilab HEP Cloud Facility.
The CMS experiment collects and analyzes large amounts of data coming from high energy particle collisions produced by the Large Hadron Collider (LHC) at CERN. This involves a huge amount of real and ...simulated data processing that needs to be handled in batch-oriented platforms. The CMS Global Pool of computing resources provide +100K dedicated CPU cores and another 50K to 100K CPU cores from opportunistic resources for these kind of tasks and even though production and event processing analysis workflows are already managed by existing tools, there is still a lack of support to submit final stage condor-like analysis jobs familiar to Tier-3 or local Computing Facilities users into these distributed resources in an integrated (with other CMS services) and friendly way. CMS Connect is a set of computing tools and services designed to augment existing services in the CMS Physics community focusing on these kind of condor analysis jobs. It is based on the CI-Connect platform developed by the Open Science Grid and uses the CMS GlideInWMS infrastructure to transparently plug CMS global grid resources into a virtual pool accessed via a single submission machine. This paper describes the specific developments and deployment of CMS Connect beyond the CI-Connect platform in order to integrate the service with CMS specific needs, including specific Site submission, accounting of jobs and automated reporting to standard CMS monitoring resources in an effortless way to their users.
The CMS Global Pool, based on HTCondor and glideinWMS, is the main computing resource provisioning system for all CMS workflows, including analysis, Monte Carlo production, and detector data ...reprocessing activities. The total resources at Tier-1 and Tier-2 grid sites pledged to CMS exceed 100,000 CPU cores, while another 50,000 to 100,000 CPU cores are available opportunistically, pushing the needs of the Global Pool to higher scales each year. These resources are becoming more diverse in their accessibility and configuration over time. Furthermore, the challenge of stably running at higher and higher scales while introducing new modes of operation such as multi-core pilots, as well as the chaotic nature of physics analysis workflows, places huge strains on the submission infrastructure. This paper details some of the most important challenges to scalability and stability that the CMS Global Pool has faced since the beginning of the LHC Run II and how they were overcome.
The connection of diverse and sometimes non-Grid enabled resource types to the CMS Global Pool, which is based on HTCondor and glideinWMS, has been a major goal of CMS. These resources range in type ...from a high-availability, low latency facility at CERN for urgent calibration studies, called the CAF, to a local user facility at the Fermilab LPC, allocation-based computing resources at NERSC and SDSC, opportunistic resources provided through the Open Science Grid, commercial clouds, and others, as well as access to opportunistic cycles on the CMS High Level Trigger farm. In addition, we have provided the capability to give priority to local users of beyond WLCG pledged resources at CMS sites. Many of the solutions employed to bring these diverse resource types into the Global Pool have common elements, while some are very specific to a particular project. This paper details some of the strategies and solutions used to access these resources through the Global Pool in a seamless manner.
Large international scientific collaborations such as the Compact Muon Solenoid (CMS) experiment at the Large Hadron Collider have traditionally addressed their data reduction and analysis needs by ...building and maintaining dedicated computational infrastructure. Emerging cloud computing services such as Amazon's Elastic Compute Cloud (EC2) offer short-term CPU and storage resources with costs based on usage. These services allow experiments to purchase computing resources as needed, without significant prior planning and without long term investments in facilities and their management. We have demonstrated that services such as EC2 can successfully be integrated into the production-computing model of CMS, and find that they work very well as worker nodes. The cost-structure and transient nature of EC2 services makes them inappropriate for some CMS production services and functions. We also found that the resources are not truely "on-demand" as limits and caps on usage are imposed. Our trial workflows allow us to make a cost comparison between EC2 resources and dedicated CMS resources at a University, and conclude that it is most cost effective to purchase dedicated resources for the "base-line" needs of experiments such as CMS. However, if the ability to use cloud computing resources is built into an experiment's software framework before demand requires their use, cloud computing resources make sense for bursting during times when spikes in usage are required.
Cloud computing is steadily gaining traction both in commercial and research worlds, and there seems to be significant potential to the HEP community as well. However, most of the tools used in the ...HEP community are tailored to the current computing model, which is based on grid computing. One such tool is glideinWMS, a pilot-based workload management system. In this paper we present both what code changes were needed to make it work in the cloud world, as well as what architectural problems we encountered and how we solved them. Benchmarks comparing grid, Magellan, and Amazon EC2 resources are also included.
The CMS Collaboration relies on 7 globally distributed Tier-1 computing centres located at large universities and national laboratories for a second custodial copy of the CMS RAW data and primary ...copy of the simulated data, data serving capacity to Tier-2 centres for analysis, and the bulk of the reprocessing and event selection capacity in the experiment. The Tier-1 sites have a challenging role in CMS because they are expected to ingest and archive data from both CERN and regional Tier-2 centres, while they export data to a global mesh of Tier-2s at rates comparable to the raw export data rate from CERN. The combined capacity of the Tier-1 centres is more than twice the resources located at CERN and efficiently utilizing this large distributed resources represents a challenge. In this article we will discuss the experience building, operating, and utilizing the CMS Tier-1 computing centres. We will summarize the facility challenges at the Tier-1s including the stable operations of CMS services, the ability to scale to large numbers of processing requests and large volumes of data, and the ability to provide custodial storage and high performance data serving. We will also present the operations experience utilizing the distributed Tier-1 centres from a distance: transferring data, submitting data serving requests, and submitting batch processing requests.
The next generation of High Energy Physics experiments are expected to generate exabytes of data---two orders of magnitude greater than the current generation. In order to reliably meet peak demands, ...facilities must either plan to provision enough resources to cover the forecasted need, or find ways to elastically expand their computational capabilities. Commercial cloud and allocation-based High Performance Computing (HPC) resources both have explicit and implicit costs that must be considered when deciding when to provision these resources, and to choose an appropriate scale. In order to support such provisioning in a manner consistent with organizational business rules and budget constraints, we have developed a modular intelligent decision support system (IDSS) to aid in the automatic provisioning of resources---spanning multiple cloud providers, multiple HPC centers, and grid computing federations.