November 2018 - Ssio-2018 - mailman3.cels.anl.gov

Fwd: Paul Messina SSIO text
by Ross, Robert B. 30 Nov '18

30 Nov '18

Hi all, I meant to send this to you all last night, but failed (wasn't watching my email completion). Apologies! Lucy: what's your take on ASCR contributions to Lustre itself? Does spending on NRE for systems get us credit for contributing to these things? PathForward (e.g., for DAOS)? Can we claim some credit for specific HEC FSIO work? Thanks, Rob Rob Begin forwarded message: From: Rob Ross <rross(a)mcs.anl.gov<mailto:[email protected]>> Subject: Paul Messina SSIO text Date: November 29, 2018 at 8:52:35 PM CST To: Mochi Mailing List <ssio-services(a)lists.mcs.anl.gov<mailto:[email protected]>> Hi all, Here's a stab at text mapping out some of the important R&D supported by ASCR in the area of storage and I/O. I've tried, but it's slanted towards things I'm familiar with, obviously. Please just send me any suggestions you might have on improvements to be more inclusive and balanced. I'm going to submit some version late tomorrow to Paul, but I'm hopeful that I will have the opportunity to continue to edit after that point. Thanks all! Rob ---

1 0

Re: [Ssio-2018] another opportunity (?)/exercise
by Grider, Gary Alan 29 Nov '18

29 Nov '18

Yes in the middle of its run for a year or so -----Original Message----- From: Ross, Robert B. <rross(a)mcs.anl.gov> Sent: Thursday, November 29, 2018 3:23 PM To: Grider, Gary Alan <ggrider(a)lanl.gov> Cc: ssio-2018(a)lists.mcs.anl.gov Subject: Re: [Ssio-2018] another opportunity (?)/exercise PLFS was funded under PDSI, yes? -- Rob > On Oct 19, 2018, at 9:46 AM, Grider, Gary Alan <ggrider(a)lanl.gov> wrote: > > I don’t know if ascr put money into hecfsio, I know nnsa and nsf and dod put money in. > If so, then there is a ton of stuff > > On 10/19/18, 8:44 AM, "ssio-2018 on behalf of Ross, Robert B." <ssio-2018-bounces(a)lists.mcs.anl.gov on behalf of rross(a)mcs.anl.gov> wrote: > > Hi all, > > Paul Messina is working on a report for Binkly related to impactful activities from ASCR during 1977-present. He's asked if I could put together something related to data, 2-4 pages, in this space. I negotiated this down to 2-4 pages on SSIO specifically, thinking that you all would want to help craft this, and that I could ask Rajeev and others for some help in identifying ASCR-supported contributions and crafting a narrative at least back into the 90s. > > He's going to poke Arie, Ian, and Kerstin for more breadth and has been talking with Jim Ahrens re: visualization and analysis, but I/we aren't responsible for those pieces. I have heard that there is some ADIOS text around already, and Paul is going to send that to me to go into this narrative. > > At this point I think the main question is this: What major activities did ASCR support that have contributed to success in the SSIO space? Then we can associate a timeline with them and write a scaffold around them, I guess. > > Specific products/outcomes that come to mind for me are mostly self-serving unfortunately: > - ROMIO > - HDF > - PVFS > - ADIOS > - PnetCDF > - IOFSL and Mochi > - Dataspaces > - data containers from LBL > - Darshan and TOKIO > - DeltaFS > > Please let me know what I'm missing. What did ASCR support out of UCSC, for example? > > Thanks, > > Rob > _______________________________________________ > ssio-2018 mailing list > ssio-2018(a)lists.mcs.anl.gov > https://lists.mcs.anl.gov/mailman/listinfo/ssio-2018 > >

1 0

Re: [Ssio-2018] another opportunity (?)/exercise
by Ross, Robert B. 29 Nov '18

29 Nov '18

PLFS was funded under PDSI, yes? -- Rob > On Oct 19, 2018, at 9:46 AM, Grider, Gary Alan <ggrider(a)lanl.gov> wrote: > > I don’t know if ascr put money into hecfsio, I know nnsa and nsf and dod put money in. > If so, then there is a ton of stuff > > On 10/19/18, 8:44 AM, "ssio-2018 on behalf of Ross, Robert B." <ssio-2018-bounces(a)lists.mcs.anl.gov on behalf of rross(a)mcs.anl.gov> wrote: > > Hi all, > > Paul Messina is working on a report for Binkly related to impactful activities from ASCR during 1977-present. He's asked if I could put together something related to data, 2-4 pages, in this space. I negotiated this down to 2-4 pages on SSIO specifically, thinking that you all would want to help craft this, and that I could ask Rajeev and others for some help in identifying ASCR-supported contributions and crafting a narrative at least back into the 90s. > > He's going to poke Arie, Ian, and Kerstin for more breadth and has been talking with Jim Ahrens re: visualization and analysis, but I/we aren't responsible for those pieces. I have heard that there is some ADIOS text around already, and Paul is going to send that to me to go into this narrative. > > At this point I think the main question is this: What major activities did ASCR support that have contributed to success in the SSIO space? Then we can associate a timeline with them and write a scaffold around them, I guess. > > Specific products/outcomes that come to mind for me are mostly self-serving unfortunately: > - ROMIO > - HDF > - PVFS > - ADIOS > - PnetCDF > - IOFSL and Mochi > - Dataspaces > - data containers from LBL > - Darshan and TOKIO > - DeltaFS > > Please let me know what I'm missing. What did ASCR support out of UCSC, for example? > > Thanks, > > Rob > _______________________________________________ > ssio-2018 mailing list > ssio-2018(a)lists.mcs.anl.gov > https://lists.mcs.anl.gov/mailman/listinfo/ssio-2018 > >

1 0

Re: [Ssio-2018] quick sync today
by Ross, Robert B. 29 Nov '18

29 Nov '18

Great, thanks! -- Rob On Nov 29, 2018, at 1:05 PM, Quincey Koziol <koziol(a)lbl.gov<mailto:[email protected]>> wrote: Hi all, I’ve finished the final bit of text for the metadata / namespaces / provenance section. Quincey On Nov 29, 2018, at 10:06 AM, Ross, Robert B. via ssio-2018 <ssio-2018(a)lists.mcs.anl.gov<mailto:[email protected]>> wrote: Hi all, Let's plan for a quick sync today to note where everyone is on their assignments. Thanks! Rob Updated: SSIO Workshop Org. Thursday, November 29, 2018 at 12:00 PM https://bluejeans.com/177339282 <93b776ecd7df4e9ea451a504bdae67dd.png> <http://bluejeans.com/> Hi there! You have updated a video meeting. Join Meeting <https://bluejeans.com/177339282/2993?src=htmlEmail> (Join from computer or phone) Phone Dial-in +1.408.317.9254<tel:+1.408.317.9254> (US (San Jose)) +1.866.226.4650<tel:+1.866.226.4650> (US Toll Free) (Global Numbers<http://bluejeans.com/numbers>) Meeting ID: 177 339 282 Moderator Passcode: 2993 Room System 199.48.152.152 or bjn.vc Meeting ID: 177 339 282 Description: Recurring call for SSIO workshop organizers to discuss progress and challenges. Want to test your video connection? https://bluejeans.com/111<https://bluejeans.com/111?ll=en> <open.gif> _______________________________________________ ssio-2018 mailing list ssio-2018(a)lists.mcs.anl.gov<mailto:[email protected]> https://lists.mcs.anl.gov/mailman/listinfo/ssio-2018

1 0

Re: [Ssio-2018] quick sync today
by Quincey Koziol 29 Nov '18

29 Nov '18

Hi all, I’ve finished the final bit of text for the metadata / namespaces / provenance section. Quincey > On Nov 29, 2018, at 10:06 AM, Ross, Robert B. via ssio-2018 <ssio-2018(a)lists.mcs.anl.gov> wrote: > > Hi all, > > Let's plan for a quick sync today to note where everyone is on their assignments. > > Thanks! > > Rob > > Updated: SSIO Workshop Org. > Thursday, November 29, 2018 at 12:00 PM > https://bluejeans.com/177339282 <https://bluejeans.com/177339282> > <93b776ecd7df4e9ea451a504bdae67dd.png> <http://bluejeans.com/> > Hi there! > You have updated a video meeting. > Join Meeting <https://bluejeans.com/177339282/2993?src=htmlEmail> > > (Join from computer or phone) > > > Phone Dial-in > +1.408.317.9254 <tel:+1.408.317.9254> (US (San Jose)) > +1.866.226.4650 <tel:+1.866.226.4650> (US Toll Free) > (Global Numbers <http://bluejeans.com/numbers>) > > Meeting ID: 177 339 282 > Moderator Passcode: 2993 > > Room System > 199.48.152.152 or bjn.vc > > Meeting ID: 177 339 282 > > > > Description: > Recurring call for SSIO workshop organizers to discuss progress and challenges. > > > Want to test your video connection? > https://bluejeans.com/111 <https://bluejeans.com/111?ll=en> > <open.gif> > > _______________________________________________ > ssio-2018 mailing list > ssio-2018(a)lists.mcs.anl.gov > https://lists.mcs.anl.gov/mailman/listinfo/ssio-2018

1 0

quick sync today
by Ross, Robert B. 29 Nov '18

29 Nov '18

Hi all, Let's plan for a quick sync today to note where everyone is on their assignments. Thanks! Rob Updated: SSIO Workshop Org. Thursday, November 29, 2018 at 12:00 PM https://bluejeans.com/177339282 [93b776ecd7df4e9ea451a504bdae67dd.png] <http://bluejeans.com/> Hi there! You have updated a video meeting. Join Meeting <https://bluejeans.com/177339282/2993?src=htmlEmail> (Join from computer or phone) Phone Dial-in +1.408.317.9254<tel:+1.408.317.9254> (US (San Jose)) +1.866.226.4650<tel:+1.866.226.4650> (US Toll Free) (Global Numbers<http://bluejeans.com/numbers>) Meeting ID: 177 339 282 Moderator Passcode: 2993 Room System 199.48.152.152 or bjn.vc Meeting ID: 177 339 282 Description: Recurring call for SSIO workshop organizers to discuss progress and challenges. Want to test your video connection? https://bluejeans.com/111<https://bluejeans.com/111?ll=en> [open.gif]

1 0

Re: [Ssio-2018] draft of streaming data section done
by Ross, Robert B. 29 Nov '18

29 Nov '18

Thanks! Comments are still helpful. -- Rob > On Nov 27, 2018, at 7:43 PM, Glenn Lockwood via ssio-2018 <ssio-2018(a)lists.mcs.anl.gov> wrote: > > Hi TEAM > > I've dropped in a complete draft of my section (5.6, Streaming Data). I'm still waiting to hear back from Amadeo on it, but I was deliberate in including specific issues that he emphasized during the discussion as well as key themes raised by Graham Heyes' presentation. > > I collected some good feedback from folks at LBNL about the pre-workshop report, and my ambition is to go through the final report draft and annotate areas that generated non-trivial comments from readers by Friday. If this will not be helpful (I realize we're late in the writing game now), I'll skip this step or save it for after tech edit. > > Glenn > > PS-hopefully this goes to the right list this time > _______________________________________________ > ssio-2018 mailing list > ssio-2018(a)lists.mcs.anl.gov > https://lists.mcs.anl.gov/mailman/listinfo/ssio-2018

1 0

Re: [Ssio-2018] draft of streaming data section done
by Nowell, Lucy 27 Nov '18

27 Nov '18

Thanks Glenn. Let’s assume that there will be no significant changes after the tech editing but only minor corrections to fix misinterpretations. Thanks, Lucy ________________________________ From: "Glenn Lockwood via ssio-2018" <ssio-2018(a)lists.mcs.anl.gov<mailto:[email protected]>> Date: Tuesday, November 27, 2018 at 8:44:09 PM To: "ssio-2018(a)lists.mcs.anl.gov" <ssio-2018(a)lists.mcs.anl.gov<mailto:[email protected]>> Subject: [Ssio-2018] draft of streaming data section done Hi TEAM I've dropped in a complete draft of my section (5.6, Streaming Data). I'm still waiting to hear back from Amadeo on it, but I was deliberate in including specific issues that he emphasized during the discussion as well as key themes raised by Graham Heyes' presentation. I collected some good feedback from folks at LBNL about the pre-workshop report, and my ambition is to go through the final report draft and annotate areas that generated non-trivial comments from readers by Friday. If this will not be helpful (I realize we're late in the writing game now), I'll skip this step or save it for after tech edit. Glenn PS-hopefully this goes to the right list this time

1 0

draft of streaming data section done
by Glenn Lockwood 27 Nov '18

27 Nov '18

Hi TEAM I've dropped in a complete draft of my section (5.6, Streaming Data). I'm still waiting to hear back from Amadeo on it, but I was deliberate in including specific issues that he emphasized during the discussion as well as key themes raised by Graham Heyes' presentation. I collected some good feedback from folks at LBNL about the pre-workshop report, and my ambition is to go through the final report draft and annotate areas that generated non-trivial comments from readers by Friday. If this will not be helpful (I realize we're late in the writing game now), I'll skip this step or save it for after tech edit. Glenn PS-hopefully this goes to the right list this time

1 0

Re: [Ssio-2018] presentation paragraphs
by Ross, Robert B. 21 Nov '18

21 Nov '18

Thanks Scott. I think there are some good ideas in here and made a few adjustments, but I tried to keep the slide high level, so omitted some details. -- Rob > On Nov 19, 2018, at 9:27 AM, Klasky, Scott A. <klasky(a)ornl.gov> wrote: > > Hi Rob, > I have suggestions for this one page for the findings. Please use whatever you guys like. > > Happy thanksgiving, > > Scott > > > -----Original Message----- > From: ssio-2018 <ssio-2018-bounces(a)lists.mcs.anl.gov> On Behalf Of Ross, Robert B. via ssio-2018 > Sent: Thursday, November 15, 2018 3:24 PM > To: ssio-2018(a)lists.mcs.anl.gov > Subject: [Ssio-2018] presentation paragraphs > > Just to put this in front of everyone, here are the summaries that I came up with. I have sent the paragraphs on specific presentations to their respective presenters for comments. The panel discussions are obviously summaries but draw from Brad's notes (thanks!). > > This is just a heads up; I will dump all this in the draft document tomorrow, after all the presenters have had a little time to suggest changes. I'd rather get changes in before I push. > > Rob > --- > Workshop Panels and Presentations > > The workshop brought together a number of experts in related fields to present their views on requirements the impact of new technologies on SSIO going forward. These presentations and panels were split across the two days of the workshop, with the first morning focusing on how application trends are creating new requirements for SSIO (application pull). > > > Application Pull > > Wes Bethel of Lawrence Berkeley National Lab presented a take on this from the perspective of the 2015 ASCR EOD (Experimental and Observational Data) Workshop [NEEDREF], reminding the attendees of the key findings from this workshop and mapping these into possible gaps in current SSIO capabilities. He noted that experimental and observational science is increasingly impeded by data lifecycle challenges: massive volumes of data being produced at unprecedented rates and in some cases with real time constraints on analysis; lack of adequate provenance and other metadata to assist in interpretation and reusability; and issues related to how to make this data available to a broad audience over potentially long timelines (e.g., years or decades). He further noted the close connection between data management and workflow in this context and a need for better coordination between these systems, and suggested that supervised learning and multi-modal data analysis will play increasingl y important roles in these science endeavors. > > [NEEDREF] 2015 ASCR Workshop on Management, Analysis, and Visualization of Experimental and Observational Data > > Graham Heyes of the Thomas Jefferson National Accelerator Facility presented a view of current and future streaming data science, augmenting the information from the prior presentation. He provided a summary of the current state of the art workflows in these fields and summarizes some shortcomings related to SSIO: the current methodology operates with close synchronization, missing opportunities for overlap and performing poorly in the face of work imbalance; use of files as an intermediate form, and batch processing of these files, forces inefficiencies in the analysis phase; and a lack of indexing and subsetting leads analysis codes to pour over more data than necessary. The end result is that these workflows can take years to complete. Moving to a streaming data model is compelling in that it could allow for more flexible design of the analysis workflow; dramatically improve the parallelism of analysis, and decouple relative rates of detectors and analysis. However, in ord er to realize this vision, new SSIO capabilities are required. First, the storage infrastructure needs to support a more stream-based model: rapid ingest of small records, strictly ordered, and easily referenced; guarantees on rates of ingest that ensure data isn't lost; and an ability to remove records over time and not suffer performance degradation from fragmentation. Additionally, esearch into multi-dimensional data stores, streaming data stores with strong performance guarantees, and methods to closely connect with high performance transports are all important for success of streaming data science going forward. > > Tom Peterka of Argonne National Laboratory presented a summary of the Workshop of the Future of Scientific Workflows [NEEDREF2], reminding the audience of nomenclature and a number of canonical workflow use cases from a variety of science fields. Summarizing the workshop's findings as relate to SSIO, he noted that (1) lagging I/O bandwidth is motivating an increased use of in situ workflows; (2) new storage technology and heterogeneity is seen as driving new SSIO work in the same way it drives new workflow advances; (3) provisioning of storage resources and services needs revisiting; (4) research is needed in how storage relates to communication in data flow, or is used in that role; (5) enhancements are needed in our ability to store and retrieve provenance, including performance data; and (6) performance models for prediction are needed to assist in workflow scheduling. > [NEEDREF2] The Future of Scientific Workflows ... April 2015 http://science.energy.gov/~/media/ascr/pdf/programdocuments/docs/workflows_… > > Jack Wells of the Oak Ridge Leadership Computing Facility (OLCF) presented a summary of data-intensive science requirements at the leadership computing facilities, with an OLCF focus. He first noted that there has been a shift since 2013 in the priorities of their users away from FLOPS and towards memory bandwidth, marking increased attention on data. Data sharing, archival capacity, and disk bandwidth were also on the radar for users, and the facilities are adjusting their systems in response. He summarized the capabilities of their new Summit system and noted successful applications in the simulation, data analysis, and AI areas. He noted that data is emerging as a common theme across many DOE offices, and that a capable storage system and scalable I/O are fundamental to success in many fields. He recognized that modeling and simulation is still a significant driver for storage system deployments, but that applications with read-heavy workloads, small writes, and heavy meta data use are a more significant portion of the portfolio than in the past. These changes drive a need for new solutions that perform better in read-optimized and random access modes, which are not the strengths of current solutions. Furthermore, continued work is needed to retain the model of a center-wide storage resource and to develop solutions that can provide a coherent view of SSIO resources even across facilities. > > Kevin Harms of the Argonne Leadership Computing Facility (ALCF) presented a summary of drivers for storage from the ALCF perspective. ALCF is similarly supporting a broad base of applications employing simulation, data, and learning approaches. New services such as Jupyter Lab and Petrel data sharing are indicative of changing requirements from applications. These tools are facilitating new methods of interfacing to the system and allowing new methods of data sharing outside the facility. ALCF is also moving in the direction of a global storage system to hold data for extended periods, and one emphasis in design of this system is strong support for legacy interfaces (e.g., POSIX). This would be coupled with dedicated storage for specific platforms that is performance focused. ALCF is attentive to advances in the object storage model as well, and is collaborating with Intel on the DAOS activity, one candidate dedicated storage solution. > > Following these presentations, a panel discussed these and other topics. Participants included the speakers as well as Evan Felix (PNNL) and Kristy Kallback-Rose (NERSC). Some important points that emerged from this discussion were the desire for alternatives to the file model including search, a recognition of the difficulty in presenting a deep storage hierarchy in a usable manner, a request for more capabilities in the area of data life cycle management, and an admission that we're no longer primarily supporting checkpoint/restart as our primary workload at some sites. > > > Technology Push > > The second morning focused on how technology developments are driving research into new directions (technology push). > > Lucy Nowell of DOE Advanced Scientific Computing Research summarized the outcomes of the 2018 ASCR Extreme Heterogeneity Workshop. This workshop was designed to identify new algorithms and software tools needed from basic research to enable facilities to support work on the SC program's grand challenge problems. Extreme heterogeneity is defined to capture massively increasing parallelism, heterogeneity of data movement and increasing levels of hierarchy, heterogeneity of performance, increasing diversity of memory and storage technologies, and increasing diversity of applications and user requirements. In this context, there is a recognition that mapping workflows onto complex hardware/software storage resources will be a challenge, that a greater degree of autonomous operation will be needed to make best use of storage and I/O resources in the face of rapidly changing workloads and operating conditions, and that new data organizations are call for in order to meet the needs of future analytic workflows, including streaming workflows. > > Gary Grider of Los Alamos National Laboratory surveyed storage and data trends likely to drive changes in SSIO solutions. He emphasized the role that economics play in the architecture of storage systems, influencing the technologies that are used to provide bandwidth, capacity, and resilience in what are typically hybrid solutions today. He noted the shrinking gap between solid-state and hard disk storage and considered whether this trend would lead to another high-level shift in architectural model. He further noted that the shift in emphasis from primarily simulation-focused to being inclusive of EOD/EOS may lead us to architectural adjustments as well, for example in response to rapid reprocessing workloads. He also suggested a revisit to the problem of resilience, and noted that traditional single-tier approaches may not provide the guarantees needed in a multi-tier context, but that solutions are being developed both inside and outside of our community. As methods to ad dress the growing volume and multi-tier nature of our storage systems, Gary suggested another look at name spaces, search, and access methods. Finally he noted that the NVME push is one of the fastest growing sectors that we have seen, and that push is bringing with it new interconnect-storage capabilities, performant user-space access methods, tight coupling to compute, and other changes that provide opportunities for the success of approaches that have not been viable in the landscape to date. > > Dan Ernst of Cray Inc. looked at the blurring lines between storage and memory in current and future systems, bringing a memory/node architecture perspective to the meeting. He discussed a number of memory technologies and considered viable configurations using multiple technologies for a simulation-focused platform, recognizing that power, bandwidth, and access density all have a significant impact on the viability of specific point solutions. The also emphasized the increasing importance of global visibility of data (across the platform) in our current era of diverse applications and coupled workflows, and noted that this naturally draws the interconnect into the picture as well, both in terms of the ability to directly access particular technologies but also the rates at which the network must communicate in specific configurations. This model, however, leads to a possible future where new global storage resources supplant the traditional role of the parallel file system a s the next layer of storage beyond application memory. > > Following these presentations, a panel consisting of Gary Grider (LANL), Kevin Harms (ALCF), Eric Pouyoul (ESnet), Dan Ernst (Cray), and Lance Evans (Cray) discussed the implications of technological change in greater depth. Alternatives to block based storage were of great interest to the panelists and audience, with no obvious consensus on whether to replace block or what to replace it with -- this appeared to be an area ripe for new ideas. The topic of offloading functionality into the NIC was also discussed as a reflection of hardware specialization that could benefit SSIO. > > _______________________________________________ > ssio-2018 mailing list > ssio-2018(a)lists.mcs.anl.gov > https://lists.mcs.anl.gov/mailman/listinfo/ssio-2018 > <SSIO 2018 Findings Rollup.pptx>

1 0