DTN on GTS Damir Regvart CARNET Jerry Sobieski

DTN on GTS Damir Regvart CARNET Jerry Sobieski NORDUNET GN 4 -3 -WP 6 -T 2 Performance Management Workshop, Zagreb, 4 -5 March 2020 www. geant. org

What is DTN (Data Transfer NODE)? • Definitions: • Hardware – physical machine and its specification • Software – the transfer tools used • Orchestration – managing the flows, possibly configuring bandwidth on demand • DTNs typically talk to other DTNs • DTNs are logically positioned with one leg in the WAN, and one leg in the campus or data center. • The datacenter side interfaces to local storage farms or high speed file systems. • The WAN side has high speed dedicated links to global core, or dynamic provisioned services across global core (e. g. NSI virtual circuits) 2 www. geant. org

DTN – all about… • All functions must be «tuned» • CPU is dedicated for data transfer • File transfer (large buffers/chunks of data), tuning TCP parameters • Logical (sequential) order in sending data • DTNs are placed out of the DMZ • DTN server has only needed services installed 3 www. geant. org

User Story for DTN • • • 4 DTN-A is situates at HPC site A. DTN-Z is situated at destination university Z Both DTN-A and DTN –Z have access to local high performance storage systems. User generates large data set called “BFDS” (Bunch of Data) at HPC A User needs to distribute this data to colleagues at University Z on another continent. User logs into DTN Portal (locally) and requests BFDS@A be transferred to DTNZ at University Z. (Authorization is assumed in this slide) DTN-A exchanges BFDS attributes with DTN-Z, and DTNs set up appropriate network and/or server/storage resources to support the transfer. DTN-A pulls data from local HPFS and pushes across network to DTNZ who spools BFDS streams to its own local HPFS disk farm User sits back and ponders his High Energy Physics problems – and does not need to worry about the mechanics of the global data motion his team requires. www. geant. org

Commercial Data Transfer Solutions • Amazon S 3 (Amazon Simple Storage Service) • platform offered by AWS that provides object storage through a web service interface. Amazon S 3 uses the same scalable storage infrastructure that Amazon. com uses to run its global e-commerce network. Amazon S 3 transfer acceleration is a combination of network and protocol based data transfer services that utilize public cloud storage resources (S 3 bucket) available inside AWS storage service. • Globus Online • File transfer service, that works on Saa. S (Software as a Service) principle. It allows users to move data between two Grid. FTP servers (like two storages) or between a Grid. FTP server and a user’s machine (Windows, Mac or Linux). • Automates the activity of managing file transfers: monitoring performance, retrying failed transfers, recovering from faults automatically whenever possible, and reporting status. • Microsoft Azure Blob Storage • Store large amounts of unstructured object data (binary or text data) that does not have a predefined data model or is not organized in a pre-defined manner. Google Cloud Platform - GCP • suite of cloud computing services that is provided by Google on the same infrastructure as internal one (Youtube, search. . . ). 5 www. geant. org

DTN work in GN 4 -3 -WP 6 • Done: • DTN User/projects survey • Geant DTN wiki page (https: //wiki. geant. org/display/DTN) • DTN tools test started on GTS platform • DTNs over GTS virtual machines • Globus Grid. FTP • Future: • Building DTN community • Continuation of DTN testing • Bare metal servers 6 www. geant. org

DTN on GTS testing (current) • 3 VMs • Grid. FTP, iperf tools testing Internet IAGW VM 1 AMST ens 3 ens 6 10. 0. 0. 1/30 GEANT 10 Gbps ens? 10. 0. 0. 2/30 VM 1 LON ens 7 10. 0. 0. 5/30 ens? 10. 0. 0. 6/30 7 VM 1 AMS 2 www. geant. org

Internet Next steps: DTN config on March 3 rd Project vlan. 1 Gbps ens 3 VM 1 LON, 1 core ens 6 10. 0. 0. 1/30 ens 7 10. 0. 0. 5/30 IAGW ens? 10. 0. 0. 2/30 VM 1 BRA, 1 core 10 Gbps GEANT VM 2 BRA, 4 core VM 2 LON, 4 core 10 Gbps ens? 10. 0. 0. 6/30 VM 2 AMS, 4 core VM 1 AMS, 1 core This config has a short hop LON->AMS (~500 km), and a long hop LON->BRA (~1600 km) Source and destination platforms tuned/optimized according to <fasterdata. url. . > 8 www. geant. org

DTN on GTS testing (future) • March – Continue VM trials over GTS. • VM based service is important – it allows flexible deployment of DTNs in/over clouds, or a [future] microservice architecture (containers? ). • Disk to Disk (SSD) - Characterize/tune I/O performance. Do we need RAID or better performance than a single SSD? • Grid. FTP - appl the DTNs actually use to move data. Multi-flow, disk to disk, over tuned servers. • Full DTN configuration: Web portal UI + multiple [virtual] I/O nodes. • Smoke test: Cross traffic – can other traffic impact performance? Either other VMs on same server, or other traffic on backbone, etc. Generate lots of side traffic and try to interfere. Implement Qo. S on DTN channels. • Analysis -> where are the limits? • April/May – Bare Metal Servers on GTS • LON -> AMS, LON -> BRA/MIL (tbd depending on server availability), • 1 CPU, 2 CPU; Multi-server (multi DTN I/O nodes) • June/July/Aug: 100 Gbps (pending approval) • Deploy 100 Gbps NICs • Test DTN configurations and tune • Possible implementation of Qo. S on links between DTNs 9 www. geant. org

SHORT demo • grid. FTP service between Amsterdam and London on GTS • Network: • 10 Gbit/s • No Qo. S (best effort) • VMs: • 4 x. CPU • Ubuntu 18. 04 LTS 10 www. geant. org

Thank you Any questions? www. geant. org © GÉANT Association on behalf of the GN 4 Phase 3 project (GN 4 -3). The research leading to these results has received funding from the European Union’s Horizon 2020 research and innovation programme under Grant Agreement No. 856726 (GN 4 -3).