...
- Round of Introductions
- Intro from Pekka, stressing importance of Nordic collab and the importance of the CTO Forum for the bigger NDN picture
- GN4-3
- Briefing on status. Shortening to 44 months
- We got good resources in WP4, WP5
- We were less successful on WP6
- We have a strong role in WP8
- Leif: the important thing is to have the foot in the door.
- LF: The project will attempt to be more dynamic;
- PN: limited resources for CNaaS
- EOSC-Nordic
- LF give status of consortium and proposal preparatium
- LJ reminds us about the dual-organisation e-Infra situation in Sweden
- Pekka reminds everybody that going forward,
- Cloud strategy intro by Lars
- 10.000 ft perspective today, identification of future joint activities, services, collaborations, actions
- Possible lay the foundation for future workshops
- SUNET perspective
- Out of scope: public cloud, HPC. SUNET focus is the long tail
- SUNET strategic focus: edge computing; keep data close to users
- The long tall is long and has really big volumes, They easily fly under the radar, and are therefore often underserved.
- The analysis tool of choice is the laptop.
- What they are asking for is "a national amazon".
- SUNET want to re-use the existing national / campus hosting capabilities.
- SUNET want to provide a local, distributed, long-tail service.
- SUNET want to integrate this with data centre networking and the CNaaS initiativ
- SUNET will build a single platform
- SUNET will own the facilities and will use a combination
- Technologies: openstack, CEPH, Leaf-spine datacentre networks, possibly include campus eScience DMZ
- Network implications: may change the network handover from "two fibres for network" to "Campus datacentre / DMZ + network"
- LJ: would like to federate with HPC (SNIC) infra, offering a single end-user portal and essentially a series of regions.
- Hybrid cloud storage: different vendors, local storage, public / commercial cloud as "regions" in the storage portal.
- Timeline: First on-prem hybrid storage deployment in Q1-2019.
- In the future, to support this, SUNET would like to primarily tender for managed on-prem. Inlcudes tendering for hw / service management (say, OpenStacjk)
- SUNET is seeing, say, OpenStack
- Compute resources are in this are so far limited - i.e., an OpenStack compute cluster next to the storage cluster. Eventually may be
- Pelle: politically, part of the attraction is the local data centre network, storage and compute node. This has made it a easier sell at the campus level.
- Will make it easy for SUNET to offer a ScienceDMZ as a managed service
- Martin: can you circulate documentation.
- LJ: whitepaper / service description document is possible. Data centre
- LF: Possible joint action on how to sell something like this
- PN: documentation, communication is on the list of things to work on
- Olaf: this makes a lot of sense. Short travelled data is important. LJ: not clear how this will evolve over time. PN: people may be moving away from massive cloud. There's a natural cycle.
- LF: down the line, integrating over HPC-like, local-long-tail and commercial cloud is an interesting topic / challenge.
- LJ: at AARNET, the approach is to make all data appear local
- UNINETT Perspective
- AL: where UNINETT is at the moment, and where we are exploring
- UNINETT focus on offering lightweight containers in a distributed setup. Have chosen K8s at the four
- Having a consistent API for the platform is key. K8s API allow the creation of an abstraction from the specifics of in-house, AWS,
- Want the system to act as a swarm, not a single large resource.
- Is looking at federation of K8s clusters.
- Want to exploit a combination of the NO higher-ed resources and commercial cloud, and have it be percieved as a single infra
- Challenges:
- Want the development experience to be more smooth
- Want better security and vulnerability detection
- L4 services require manual intervention
- Accounting, centralized is a real challenge, and is needed for a comprehensive service offering
- Today mostly offering services that do not have a lot of persistent storage
- Currently have two clusters running. One running on top of a IaaS cluster, one running on a UNINETT Sigma cluster in Tromsø
- LJ: what kind of components are offered? OK, AL: core K8s, orchestration, a few extra services. More complex services not there yet
- UH-Sky project is ending this year. Final stages of program lives with UH-Sky
- One UH-Sky experience is that a cloudification initiative is needed, to help community take up cloud services
- Question: should UNINETT act more as a single point of contact for infrastructure, or a be a broker. Similar questions exist w FUNET, SUNET
- AL: users are interested in the service, not the hardware or platform. It's really software-as-a-service we want to do.
- UNINETT Sigma has built a GPU cluster that can be used for these types of services
- UNINETT: customers want abstractions. Pekka: FI users are pushing for same thing. JM: there's a Dutch project that is now in production.
- Lots of discussion of tendering and the implications of IaaS vs on-prem
- Question:
- should uninett be a cluster federation provider
- What kind of ecosystem around the cluster can UNINETT provide (T&I, logging, monitoring, accounting, app-store, ...)
- LF: PRP / NRP in the US have chosen K8s, and a doing a lot with federating smaller clusters etc. We can connect with them and share experience.
- Olav: UNINETT is further developing the Application-as-a-Service / data-analytics-as-a-Service as a portal. Now allows multiple cluster of execution and a hybrid execution platform.
- Being used also for classes and universities.
- UNINETT has developed a proof-of-concept, that is now being run by Sigma as a trial service.
- Development is being shared as part of the GLEMMA project (with CSC, NeiC). Collaboration with CERN. Hope to inject it into GN4-3 WP4T3.
- Next step will be to integrate with IaaS backends.
- Pekka: how are we handling accounting / invoicing. OK: in development
- LF: so UNINETT has two layers of service (for long tail): a K8S-based federated cluster infra service, and a data analytics as a service at a layer above.
- OK: next evolution step will be to offer machine leaning services
- OK: something like this is a big thing for a single NREN to do, so collaborating would be useful
- OK, JM: there's items in the GN4.-3 Work Program that could support this
- OK: you can log in with federated login and try this. Maybe.
- FUNET - CSC cloud strategy
- CSC cloud roadmap developed some years ago
- Compute services strategy, CSC compute services portfolio
- A range of *ssS service models for cloud computing provided by CSC
- Includes a container-based K8s infra service platform, similar to UNINETT service, to go into full service soon
- CSC offers a resonable-sized K8s cluster, and one for sensitive data services
- There general use cluster is a slice of the HPC platform. The sensitive data cluster is separate.
- Pekka: CSC is in tender for new HPC platform. So far unknown if the new platform will also be used for cloud services, or if that will be separate.
- Pekka: billing / accounting is a challenge. This is not built in. LJ: the SUNET service have the accounting built in from the start.
- LF: the European ideas for Cloud Coins may be something to look into
- OK: accounting should go hand-in-hand with monitoring, to understand real availability
- FUNET service evolution
- FUNET Kampus (CNaaS) - in pilot
- FUNET Capacity Services - backup in pilot, data centre hosting in planning. Similar to Swedish ideas
- When FUNET CNaaS was initiated, it was envisioned that it would be the small campuses wanting this. Turns out the large universities are signing up.
- It is so far not clear if this will be mostly focused on the border management, or will evolve into campus LAN / wifi management.
- PN: SUNET is planning to run one campus network in 2019, in order to test the toolset development.
- Strategic position
- DeIC
- DeIC is doing HPC; those services are not really related to cloud services
- DeIC is also offering the GÉANT IaaS services. However, it has very little uptake
- a single shared service based on nextCloud - sciencedata.dk
- DeIC would like to do a service similar to sciencedata.dk for computing - an open, free VM service. No backing from campuses so far.
- At a single Danish campus, individual researchers are procuring VM with their individual credit cards in fairly large quantities, probably above EC procurement limits
- DeIC is working with NeIC in Nordic cloud service project.
- VN, LF, JM: we have a challenge w/ ensuring communications lines for research support, and not just the business process view.
- RHnet
- Interested in the services that are being developed
- Iceland government has contracted with MS for a cloud service for all public sector including education & research. As a result, MS has a lock-in on the Icelandic R&E cloud market.
- LJ: the challenge of vendor lock-in for basic collaboration tools is massive at the moment.
- PN: we (NRENs) should be careful promoting these services that can lead to lock-in
- MO: the above is the reason there's no RHnet cloud strategy
- Cloud Service Delivery
- GÉANT cloud service delivery framework - it's there, how do the Nordics relate to it
- Consumption of commercial cloud services may be direct, but the responsibility is still with the institutions
- Trends
- our community likes hyperscale / global providers
- out community appear to like SaaS
- Hyperscale providers are getting bigger,
- R&E is falling behind on adoption, our fraction of revenue with providers with them is declining
- Introduction to the GÉANT activities
- Leverage joint effort, scale over many NRENs, save (human) resources
- pan-European procurement
- light-weight collaboration with providers
- Jan M want to turn this in to a repeatable process
- GÉANT IaaS
- Provider / service catalogue ready
- 2018 pan-European revenue is about 7 M€; not a lot of that from the Nordic region
- Currently 300 consuming institutions, 18 of 36 countries
- There's fast uptake in some countries, in others not.
- In Sweden, Finland, campus legal departments is often the blocker for uptake
- large consumption happens in Spain, Ireland, Netherlands. These countries had the business processes in place already
- Most of the consumption is Azure
- Next steps
- continue IaaS
- HPC-focus in OCRE. Tender in September 2019. Will try to do the tender so it's not just for HPC, but essentially v2 of the IaaS tender
- Private cloud, possible using K8s containers
- What will the Nordic do
- JM: if we collaborate for exploiting the delivery framework, we can achieve more
- JM: Do DK, SE, FI, IS want to build on the Nordic position / reputation, or just be "any other NREN"
- HEAnet, SURFnet, DFN are showing ways to benefit from, and to let members benefit from, the IaaS framework
- Ideas for Nordic way forward
- create a coordinated nordic process for demand aggregation
- Coordinated input to requirements for next IaaS iteration
- Martin: given the disappointing uptake, what are the chances of getting a good deal in the next iteration; JM: uptake is growing, vendors are so far reasonably happy. Vendors w good local representation have an advantage
- Next?
- How do we participate in the next GÉANT IaaS. Valter: do we do GÉANT or Nordic? PN: we should do an independent workshop. Pekka: we need a champion, NDN should coordinate
- Qvist: maybe people need more platform or application or HPC as a service, and less plain Infra-as-a-Service
- LF; other topics to take away: 1) data centre / CNassS / SUNET distributed cloud storage services, 2) K8s-based federated community clouds 3) Participation in GÉANT IaaS Framework
- Qvist: why are people so afraid of firewall-as-a-service?
- Steps:
- Learn about what we are doing
- Learn if Nordic NRENs have joint objectives and / or a joint strategy
- identify how NORDUnet can support that
- AL: strategy must also look at how to deal with public cloud providers, Office360, and how institutions cope with that.
...