August 2012 - Asg - mailman3.cels.anl.gov

Wiki update: Reporting
by Wiki update: Asg-private: Gcdanie 03 Aug '12

03 Aug '12

3 2

Wiki update: File:Mpo-40001110000-deliverables-update-07-2012.pdf
by Wiki update: Asg-private: Dkimpe 03 Aug '12

03 Aug '12

1 0

Wiki update: File:Mpo-40001110000-deliverables-update-07-2012.pdf
by Wiki update: Asg-private: Dkimpe 03 Aug '12

03 Aug '12

1 0

Wiki update: Reporting
by Wiki update: Asg-private: Dkimpe 03 Aug '12

03 Aug '12

1 0

Wiki update: File:Mpo-40001110000-report-input-07-2012.pdf
by Wiki update: Asg-private: Dkimpe 03 Aug '12

03 Aug '12

1 0

5-year-plan base document
by Ward, Lee 03 Aug '12

03 Aug '12

All, This is supposed to be a formatted roll-up of all our input, so far. Please check it to make sure I have not missed something that you sent or incorporated a wrong version. I was supposed to generate the software artifacts document this evening but it's pushing midnight and I'm getting loopy. It's first up on my list tomorrow morning :( Yes, I am aware that my word processing skills leave a lot to be desired. Along those lines, if anyone knows how to undo the transparent japanese language setting I somehow got in there I would really appreciate it. Same goes for the "plain text + arial 12 pt" setting that is ubiquitous for all the body text. It's plain, I should not be allowed to do things like this. Hopefully, as a last pass tomorrow someone who actually knows what they are doing can fix my stupidities. Sorry, I really am trying. --Lee

3 3

Wiki update: File:ASG July 2012 Monthly Report.pdf
by Wiki update: Asg-private: Gcdanie 02 Aug '12

02 Aug '12

1 0

Re: [ASG] Suggested milestones
by Dries Kimpe 02 Aug '12

02 Aug '12

* Richard Brooks <rrb(a)acm.org> [2012-08-01 13:36:07]: > > * [Clemson/Brooks] Development of tools for analytical > > verification of protocols for deadlock, livelock, performance, and > > information leakage by leveraging AESOP primitives. > > --> Still not sure about this one. Do we have any more detail on > > how this would actually work before committing to delivering this? > How does it sound to do planning here and implementation in > the out years? It's not about the timing; It's just that I don't currently see how aesop primitives make this task easier/possible (that probably says more about me than about the deliverable). Maybe leave aesop out of it, but keep the rest? That would give you the flexibility to investigate other options if needed. > > * [Clemson/Brooks] Tools for identifying and mitigating denial of > > service incidents. > > ----> Denial of service and related events are basically accounting > > for 50% of yr3, 33% of yr 4 and 50% of yr 5 (with yr 4 and yr 5 > > being the same deliverable). Can we really justify so much effort > > to this one topic? > > Is it an issue in similar storage systems? For example, do > > lustre/gpfs/... have anything in this area and if not, are they > > looking into adding this? > > Given the intended deployment of the storage system, there might be > > other areas that could provide more benefit? > Do you have suggestions for anything you might find more useful? > That would be welcome. Coming up with 5 year plans with a one day > turn around is challenging. > Ideas from you are very welcome at this point.... Unfortunately, no. :( The only thing I can come up with now is to investigate and implement the authorization for the other frontends we would support? (i.e. for posix for example), but not sure if this makes a good deliverable. (It is something we need though). Ideas/suggestions from other people? Dries

5 11

5 year plan V2
by Ward, Lee 02 Aug '12

02 Aug '12

All, Attached, please find what I believe to be a roll-up of input you have delivered so far. Please check to make sure I haven't deleted anything or otherwise screwed up. Still need input from Jill, UAB, and TAMU. I think I need that by close-of-business tomorrow, I'll need much of Thursday to try to organize this and format it. Then, Friday, I think we go for the down-edits and consensus from the group for the final document we want to turn in. I hope you saw the note from Brad? He tells us that we really only need to turn in a list of software artifacts. I'll use Thursday to try to extract that from the information you have supplied and we will have to work that on Friday as well. Rob never withdrew his request that we accomplish this, though, so I intend to finish what we have started as well. Given the two different assignments it seems easier to just do both. --Lee

3 2

Re: [ASG] [EXTERNAL] Re: A question about the model
by Ward, Lee 02 Aug '12

02 Aug '12

I think the answer is yes. I also think that is not optimal, we need a way to tell the client which is which. Better , maybe, would be to tell the client whether they can access an authoritative copy "here" or, if not "here", where. Unless I'm missing something? Seems to me the question is only relevant when the caller is worried about coherence, no? Are there other use cases? --Lee -----Original Message----- From: Rob Ross [rross(a)mcs.anl.gov<mailto:[email protected]>] Sent: Monday, July 23, 2012 02:03 PM Mountain Standard Time To: Settlemyer, Bradley W. Cc: Subject: [EXTERNAL] Re: [ASG] A question about the model My take on this is that the language says that objects *can* span servers, not that they must :). The API does not expose any way for a client to ask for distribution in this way or to know if this has been done, so it is left to the implementations to decide how/whether to distribute data for a given object across multiple servers? Rob On Jul 23, 2012, at 2:42 PM, Settlemyer, Bradley W. wrote: > Okay, so I am mistaken on when the change occurred. So that's at least > one mistake. I hope I also misunderstand what the note text means. > > How do you name objects so that they span more than one server (while also > ensuring no other server also uses a name that causes a collision)? You > can tack the server onto the object name, though how you pick which server > to tack on is pretty tricky. I'm getting lost as I try to work out how > you can uniquely name objects that span servers without ensuring that name > is unique. At the very least, this will not be a simple scheme, and at > worst, its going to require PAXOS . . . > > Cheers, > Brad > > > On 7/23/12 3:34 PM, "Dries Kimpe" <dkimpe(a)mcs.anl.gov> wrote: > >> >> * Settlemyer, Bradley W. <settlemyerbw(a)ornl.gov> [2012-07-23 15:22:39]: >> >>> Hello >> >>> I assume a model conference call occurred without my knowledge, >>> because I see the following line in the model document: >> >> >>> Note >> >>> Need to explain distribution. Containers, objects and forks can be >>> distributed over multiple servers. >> >> >>> That surprised me! Objects can span servers!? Forks can span servers?! >>> Maybe you are using a different definition of server than I think you >>> are using? These all seem like bad ideas to me for a variety of reasons >>> ‹ but at the very least, naming is problematic in this model. Am I >>> mistaken? >> >> That has been in there since 2012-04-04 15:38:43 -0500, >> definitely not a recent change. >> >> It looks like this was added when we introduced explicit locations to >> the API. >> >> Dries > > _______________________________________________ > ASG mailing list > ASG(a)lists.mcs.anl.gov > https://lists.mcs.anl.gov/mailman/listinfo/asg _______________________________________________ ASG mailing list ASG(a)lists.mcs.anl.gov https://lists.mcs.anl.gov/mailman/listinfo/asg

4 9