Theo Lynn

Theo Lynn

Theo Lynn is Professor of Digital Business at DCU Business School and PI of the Irish Centre for Cloud Computing and Commerce. He is also a PI on the EU Horizon 2020 RECAP and CloudLightning projects.

Location Ireland

Activity

  • Bingo!

  • I am not so sure they will be replaced but the cloud will take over some workloads

  • CloudLightning ; )

  • As above, do you not think that the cloud is useful for certain types of HPC workloads? - loosely coupled or pleasingly parallelisable ones but for the "true HPC" workloads it probably doesn't work that well.

  • True. However some communications technologies are promising and may achieve step changes in data transfer which may make "true hpc" in the cloud more feasible in the medium term....however the definition of high performance is a movable feast so customer expectations will also be higher by then.

  • It is easy to fall in to that trap. Many large scale HPC workloads simply don't work in the cloud./

  • Yes, a hybrid solution is a possibility. But also as telecommunications technology improve, it is possible we could see HPC in the Cloud for larger workloads a reality.

    I think the reality is that the cloud is useful for certain types of HPC workloads - loosely coupled or pleasingly parallelisable ones but for the "true HPC" workloads it probably doesn't...

  • But this course argues that you can democratise HPC and allow smaller organisations rent specialised infrastructure without the CAPEX of traditional HPC? Do you not see the potentialities there for HPC in the cloud?

  • I think the issue will be whether cloud will replace HPC?

    Governments are more likely to use hybrid solutions. No?

  • @SAMUELOLAGOKE some of these issues exist independent of the cloud though? Maintaining and upgrading is easier in the cloud.

  • Hi Christine. It depends at what level you are using the cloud. For example, IBM provide bare metal as a service

  • @Dell'OsteVincenzo @BillLaing Security is a contentious issue - some say the cloud is safer, some say it is inherently not as safe as on-premise. To some extent it comes down to trust and control. I tend to believe that major hyperscale cloud service providers are much better positioned than the overwhelming majority of enterprises at security. However,...

  • It can be but remember at a macro-level, all these data centres consume a lot of electricity which is also bad for the environment!

  • Hi Daniel. There are readings at each chapter and we will be emailing everyone to give you access to an open access book that complements the course. If you want to get ahead - you can download at https://link.springer.com/book/10.1007%2F978-3-319-76038-4

  • Great. Dell EMC is forging ahead in this area - https://www.dellemc.com/en-ie/solutions/high-performance-computing/index.htm

    Welcome!

  • Great. We will be focussing more on HPC in the Cloud, which is quite niche, but welcome!

  • I definitely think things have improved. This course doesn't particularly look at security but security is a major factor in cloud adoption and reticence.

  • Well I'm glad you are doing the course then. As Francis Bacon says:

    "There is nothing makes a man suspect much, more than to know little, and therefore men should remedy suspicion by procuring to know more, and not keep their suspicions in smother."

  • Hi Joseph. This may be a bit advanced but welcome!

  • Hi Suwash. Wow, interesting field. Hopefully there will be some things you can apply from this course in your research.

  • Hi John. Welcome. One of the use cases we will look at briefly and use as a reference point is OPM - https://opm-project.org/ and the Upscaling module specifically.

  • Hi Saida. Interesting. While this course doesn't really deal with that topic. We are looking to apply the CloudLightning Simulator to telecommunications use cases in the Horizon 2020 RECAP project. More info at https://recap-project.eu/

  • Some of the contributors in this course are from CERTH in Thessaloniki and DUTH. Weclome!

  • @WalidShaari I think this largely depends on the workload and configuration of the HPC cluster in the cloud

  • Is this a function of how much you are willing to pay for the QoS you need?

  • What if its a private cloud? Is security still an issue?

  • I like how you are are categorising here. The question is what does HHPC and LHPC mean? Is it loosely coupled vs tightly coupled? Its a movable feast of course.

  • We have that today. Rendering is the low hanging fruit....what about the truly high performance computing workloads...high performance is a moving target? Even in rendering as we develop devices for displaying higher resolution 3 dimensional imaging, we need higher powered processor units etc...

  • Of course, they are hard to answer - they are designed to make you think! I like your answer though - balanced and pragmatic. I think the security issue can be a huge distractor - in some cases, yes security is an issue e.g. big pharma...in other cases maybe not so much.

  • Hi Robin. I agree. The ability to scale up and down (ideally automatically/autonomically) is a key advantage particularly for SMEs.

  • Hi Katherine. The democratisation of enterprise computing and in this case high performance computing which may have been costly, inconvenient to access, or difficult to use without specialist training is a key advantage of cloud computing. I truly believe HPC in the Cloud, in particular, can play a huge role in accelerating innovation by unleashing computer...

  • I agree, Patrick. Speed to market and speed of deployment is incredibly important and can be a significant source of competitive advantage.

  • Hi Katherine. DFEs in particular may be interesting for you - they are widely used in financial services and in this course, we look at their use for genomics. These are typical high performance data analytics use cases. Welcome.

  • Welcome Kat. This course is very HPC in the cloud focussed....it may not hit all your requirements but welcome!

  • Hi Magnus. It is great that you have some experience with AWS - they are pushing the boundaries in cloud and HPC through what they term "high throughput computing" and indeed are offering CPUs, GPUs and FPGAs in the cloud so this course should provide some interesting insights for you. Welcome.

  • Hi Sandeep. You seem to have nearly an ideal background for this course. We'll be touching all those topics - HPC, cloud, containerisation etc throughout the course. Stick with it!

  • Green computing is (can be) something different than the cloud. Indeed we will discuss some debates about how green cloud data centers really are. In Week 1, we should deal with some of these issues. Welcome, Ssemujju!

  • Hi Mark. Welcome - hopefully we'll meet your research and educational requirements!

  • Hi Ananth - we'll be talking about HPC a lot - most in the cloud but it should give you a lot of insights about HPC.

  • Welcome Pete. We'll deal with the cloud straight away in Week 1!

  • Hi Valerie. Hopefully you'll learn some interesting things about the cloud in Week 1. Week 2 onwards gets a bit technical but hopefully you'll get something from the course. Welcome!

  • Hi Srinivas. I think we'll answer some of your questions in Week 1. Welcome.

  • Hi Simon. It sounds like you are at the coalface already. Hopefully, you will find some insights in this course

  • Hi Florrie. Welcome. Certainly, I believe that HPC in the Cloud will transform healthcare by making genomics and dna testing affordable for every clinic and hospital.

  • Failte Emma! We will be looking at some specific genomics use cases later. Certainly it is an area where the cloud can be used to process data for your use case.

  • Hi Ukpong. We'll be answering that very question in this first lesson!

  • Failte Carla.

  • Hi Fred. This should be very different from your normal field of study....definitely a jump in the deep end of the pool! Good luck!

  • Ok Yukiko. These are all very different. I think the first few lessons should clarify the differences between HPC, the cloud etc.

  • Hi Alberto. Certainly this will bring you down and different path which overlaps your previous study. Welcome.

  • Great Amitabh. You might learn a bit about supercomputing and much more about HPC and the Cloud.

  • Welcome Suresh. Lots of aerospace use cases for HPC and HPC in the cloud.

  • Hi Leo. Intel worked on our project and obviously we are looking at a variety of processors throughout the course so this is probably relevant to your industry in particular.

  • Welcome Agostino! Hopefully we'll give some insights.

  • Wow. That's a niche area but growing. Very interesting.

  • Hi Richard. This course should give you some refresher knowledge but also some new insights. Good having you on board.

  • Hi Amir. There is definitely some overlap in the ideas here and there should be interesting perspectives on simulation etc. Welcome.

  • Great. Some of this will provide an overview - some is probably more fine grained or in the future but should all be relevant. Welcome.

  • Hi Marci. I have no doubt in the future HPC in the cloud will be used for dna testing and genomics in clinics and hospitals throughout the world. Welcome.

  • Good use case. Not one we've looked at but very close to weather simulation.

  • Much of the course could be relevant to the telecoms sector particularly as the telecoms companies look for new markets.

  • Halo! Am I saying that right in bislama. Anyhow, you are welcome to the course.

  • @DavidMackereth Quantum computing...sure, if you can get over the cost, temperature requirements, programming hurdles, scaling etc. Why not!

  • @DavidMackereth Yes, a colleague of mine - Dr Johnny Walker - was involved in that business in Australia - https://www.siliconrepublic.com/start-ups/johnny-walker-interview

  • You can use the cloud to process data in less time.

    I think you need to tease this out more. Is there any scenarios where cloud computing for HPC can replace HPC? Of course. All cases?

  • @DaveAllcock I think your answer is probably the right one. Over a long enough timeline, anything is possible. Resolving the telecommunications/networking issues may change our understanding of what is possible

  • I agree. The ease (cost and speed) of data transfer is a key bottleneck...not easy to resolve either for the larger, more complex workloads and applications

  • I agree. But also data transfer, data management etc.

    If there is a step change in telecomms technology then HPC could be replaced but realistically we are years, if not decades off, being able to have cloud-based HPC that competes against supercomputers for the more complex workloads

  • Good post

  • Cloud is a specific and defined thing - http://nvlpubs.nist.gov/nistpubs/Legacy/SP/nistspecialpublication800-145.pdf

    Indeed, definitionally HPC may be more nebulous and a moving target. What is high performance?

    A shared national supercomputer is probably not a cloud because it does not meet the tests of a cloud

    On-demand self-service
    Broad network...

  • It can at single point level but can it as a whole?

  • That's a good insight.

    We may have to change or arrive at new definitions of performance.

  • Just dropping some random workload on a cloud will most likely fail. No doubt.

    I think you answer the question by asking a question. The answer is no or rather not in the near future. There will always be a increasingly greater definition of "high performance" and the cloud will most likely lag behind this.

    However, innovations in telecommunications may...

  • I agree. HPC in the Cloud can democratise HPC and allow individuals or SMEs compete and innovate in a way that is currently inaccessible or prohibitively expensive.

  • @ROCKSONANKOMAH Again....FPGAs and other resources will dealt with later in the course but will the cloud replace HPC?

  • While this link referenced HPC...does it answer the question?

  • You know, I am not sure it is not expensive. In some cases it could be.

    I think for some applications and workloads correctly architected for the cloud, the cloud offers tangible (cost savings, performance etc) and intangible benefits. But not in all cases!

  • Ha! I think you point to another valid area for examination.

    There is a need for greater scrutiny of decision-making regarding migration to the cloud...not just total cost of ownership but return on investment. This implies greater focus on intangible benefits. Accountants tend to focus on cost. Engineers more on performance. We need more focus (and...

  • I think in this context, you are right....size matters.

    Larger companies may have more skin in the game both in terms of capital expenditure, need for dedicated resource etc etc.

    Smaller companies currently may not have the option of larger-scale HPC and the flexibility and economic advantages of the cloud may outweigh other considerations.

  • I agree. The cloud needs to adapt for HPC. Similarly, maybe HPC (vendors and services) need to change to offer more cloud-like services.

  • I am not sure "cloud computing" is quite marketing hype. There has been different flavours of distributed computing but to be in the cloud is a very specific thing:

    On-demand self-service
    Broad network access
    Resource pooling
    Rapid elasticity
    Measured Service

    Webmail, FTP etc did not necessarily meet all of the above criteria.

    Here, HPC in the...

  • @DavidMackereth Models for protection of electronic personal health records already accommodate many of the use cases in medical practice and address your queries. I think cloud and IT, in general, has great potential to leverage GPs further but also other geographically disperse health specialists e.g. radiographers or medical specialists. Alliance Medical...

  • @HughNicklin I agree Hugh. This is particularly relevant to HPC in the cloud where entry level machines are prohibitively expensive. I think it allows HPC users trial software, models or ideas at smaller scale before committing to a full roll out but also allows for surge workloads or overflows where existing capacity isn't available.

    I don't think the...

  • That is a pre-requisite or drawback but not part of a strategy

  • @SanthirasegaramBalenthiran I disagree with you. Online scanning and outsourced interpretation of xrays and other imaging outputs is transforming medical practice particularly in remote regions.

  • @DavidFoster There are certainly downsides in the cloud model and in HPC in the Cloud also e.g. minimal interference, lack of transparency on location or control of data/workloads and as you mention transfer of control to the vendor exposes companies to vendor-related vulnerabilities..

  • Do you believe the marketing hype?

    Reduce energy footprint - maybe on a unit level but in aggregate cloud computing probably has a greater energy footprint

  • Certainly security is different...once could argue that you can leverage the better security of specialist vendors...on the other hand, as you correctly, point out you expose yourself to greater risk....

    I think less of benefit or unknown at this point but a good contribution to the wider debate

  • I think some of these aren't necessarily cloud specific and could be covered in other areas but certainly are benefits

  • Theo Lynn made a comment

    Thanks everyone who took the time to introduce yourselves. We have a varied group of learners with different motivations and interests. Hopefully you'll all find something of interest over the coming weeks.

  • Hi Gareth. We've been looking at your project - we particularly liked some of your outreach videos. I think you'll find some interesting content over the coming weeks.

  • Hi Nigel. Week 6 will be particularly interesting to you - we discuss how a new cloud architecture, CloudLightning, addresses the challenges of HPC in the cloud

  • Climate change simulations are good example of a HPC workload - it would be an interesting challenge for HPC in the Cloud....

  • @HughNicklin @SeanDelaney so Prof John Morrison, the Coordinator, was a founder of Grid Computing Ireland and has been at the forefront of HPC, grid and distributed computing for some time. Obviously both DCU and UCC are consortium members of ICHEC. You'll also see various contributors have interacted with PRACE e.g. Anne Elster (NTNU) etc. The big challenge...

  • Welcome Harry! As an occasional entrepreneur myself, it's great to see you on the course. Some of the outputs we'll be discussing are open source so opportunities abound!

  • For emerging markets like Nigeria, I think HPC in the cloud will allow researchers and industry leapfrog or get a foothold without the huge capital expenditure needed for supercomputing infrastructure.

  • @FrancisPham this course should be of interest then...we looked at genomics as a use case- obviously HPC plays a huge role in healthcare and pharamceutical R&D in particular. Any way of democratising HPC is going be good for the healthcare sector.

  • Hi Ian. We liked your video on what HPC can do for you! We did some briefings in London earlier in January and met quite a number of people from the UK HPC community. There was some reticence about HPC in the cloud - we see it as complementary for loosely coupled workloads (of which there are a lot!) - thus freeing up resources in the centralised supercomputer...

  • Week 1 should help with the introduction. Welcome on board!

  • @HerbertMehlhose this looks a good pick for you...it's going to look at a lot of the challenges facing HPC and how they might be addressed by the cloud through heterogeneous resources and new approaches to resource management