ࡱ> AR@{r( / 0LDArialЃ 0Ѓ0hh0 0@ .  @n?" dd@  @@`` P>      0AA@3ʚ;ʚ;g4?d?d= 04ppp@ <4dddd4w 0h0 80___PPT10  LEAD-VGrADSDay 2 AMDiscussion TopicsvgES Performance Model Distributed control algorithms and how to handle workflow engines Implementation details of middle layer}vgESRefer to Yang-Suk s slides Nested slot  Can specify for a cluster or tight bag and applies to all children  hierarchical What needs to be specified to scheduler Hostname/<scheduler> allocationId NoOfProcessors Input/Output CPU time, wall time WS-GRAM RSL example How will vgES handle GT security? As far as possible keep the LEAD GRAM and vgLaunch similar BPSPtPStb A@   >Are we combining allocations? Options We have to do pre-submissions to the queue Optimization to reuse one allocations We manage our local pool with our own job manager Looks to PBS as one queue DVC needs to be started on every single node Globus gatekeeper on every single compute node? Seems like a lot of overheadj'ZQZ2ZxZZ'Q2x   nWe can submit job to DVC or just use GRAM. First step is to submit to a batch queue Next step is to see the optimization problem TeraGrid logistics will run our own MDS What does this mean to others batch queue prediction Probability number At that time, by that time are two different qns what needs to be supported in the language Add/remove nodes scheduler,X\ ,X  \ ,H yNeed a resource for 1 hr for time t Hard to do unless you use advanced reservations Need a resource for 1 hr by time t Can we have two separate language constructs? We propose we should Are we going to have slots for SC demo Looks like we need it to be able to support deadline Yang-Suk says we will have fixed slot Need to fix language a little bit Some stuff to support slot$0Q'[=$0Q'[  =[ QLong term question: How will vgES support application specific activities? Will callback that vgES have help? Pegasus represents data as tasks in workflow This will not work for LEAD because it has streaming data To play some interesting games we need some state object that can be used to reserve slots Not immediately but we need thisKZ#Z-Z;Z[Z!ZK#-;[! ,=Dynamic aspects of LEADSome slides (in Dennis stack from Houston meeting) (document on dynamic workflows) Motivation for nested workflows, control weather events control radars Dynamic workflows (still WIP in LEAD) Conditional and loops in workflows (next task is unknown) Can have weights for the possibilities This with constraints will be very tricky and interesting What do u do in the case where someone is not happy with your response and is willing to negotiate How does someone accept drift?  ZZ&ZZcZ Z &c   Real-time prediction case need the schedule needs to be automated right now user driven How do we make sure we don t promise the same resources to more than one person? In general when failures happen, what do we do? Not making the constraint is just one of it `B.B./      0` 33` Sf3f` 33g` f` www3PP` ZXdbmo` \ғ3y`Ӣ` 3f3ff` 3f3FKf` hk]wwwfܹ` ff>>\`Y{ff` R>&- {p_/̴>?" dd@,|?" dd@   " @ ` n?" dd@   @@``PR    @ ` ` p>> f(    6`r  `}  T Click to edit Master title style! !  0Lu  `  RClick to edit Master text styles Second level Third level Fourth level Fifth level!     S  0| ^ `  >*  0L ^   @*  0 ^ `  @*H  0޽h ? 3380___PPT10.*c< Default Design 0 zr0  (     0( P    P*    0     R*  d  c $ ?    0p   0  RClick to edit Master text styles Second level Third level Fourth level Fifth level!     S  6 _P   P*    60 _   R*  H  0޽h ? 3380___PPT10.*c`_M$ 0  $(  r  S tʊ>  r  S Њ `    H  0޽h ? 3380___PPT10.*c<$  0 P$(  r  S ,Š `}   r  S Ê `  H  0޽h ? 3380___PPT10.*c$N$  0 p$(  r  S 4t `}   r  S w `  H  0޽h ? 3380___PPT10.cЦ  0 $(  $r $ S $ `  H $ 0޽h ? 3380___PPT10.cpg  0 ,(  ,r , S P `  H , 0޽h ? 3380___PPT10.c,  0 4(  4r 4 S pP `  H 4 0޽h ? 3380___PPT10.c\x  0 <(  <r < S J `  H < 0޽h ? 3380___PPT10.cp8$  0 D$(  Dr D S   `}   r D S ` `  H D 0޽h ? 3380___PPT10.c0a9  0 0L(  Lr L S (  `  H L 0޽h ? 3380___PPT10.cPv  0 @(  X  C       S   0    H  0޽h ? 3380___PPT10.*c[aM  0 `(  X  C       S :  0    H  0޽h ? 3380___PPT10.*c0Ɓ  0  (   X  C       S @  0    H  0޽h ? 3380___PPT10.cD  0 ((  (X ( C      ( S   0    H ( 0޽h ? 3380___PPT10.cN  0 0(  0X 0 C      0 S   0    H 0 0޽h ? 3380___PPT10.cQ  0 8(  8X 8 C      8 S O  0    H 8 0޽h ? 3380___PPT10.c@   0 @(  @X @ C      @ S    0    H @ 0޽h ? 3380___PPT10.cpam  0  H(  HX H C      H S 3  0    H H 0޽h ? 3380___PPT10.ce  0 @P(  PX P C      P S t*  0    H P 0޽h ? 3380___PPT10.c%rXPz(+&%T-R<f>/z@1B^3D5F6Ht8J:LO1( / 0LDArialЃTopicsvgESSlide 4Slide 5Slide 6Slide 7Dynamic aspects of LEADSlide 9  Fonts UsedDesign Template Slide Titles 73Microsoft Office PowerPoint@ 3_0UNCUNCg    y--$xx--'@Arial@n?" dd@  @@`` P>      0AA@3ʚ;ʚ;g4?d?d= 04ppp@ <4dddd4w 0h0 80___PPT10  LEAD-VGrADSDay 2 AMDiscussion TopicsvgES Performance Model Distributed control algorithms and how to handle workflow engines Implementation details of middle layer}vgESRefer to Yang-Suk s slides Nested slot  Can specify for a cluster or tight bag and applies to all children  hierarchical What needs to be specified to scheduler Hostname/<scheduler> allocationId NoOfProcessors Input/Output CPU time, wall time WS-GRAM RSL example How will vgES handle GT security? As far as possible keep the LEAD GRAM and vgLaunch similar BPSPtPStb A@   >Are we combining allocations? Options We have to do pre-submissions to the queue Optimization to reuse one allocations We manage our local pool with our own job manager Looks to PBS as one queue DVC needs to be started on every single node Globus gatekeeper on every single compute node? Seems like a lot of overheadj'ZQZ2ZxZZ'Q2x   nWe can submit job to DVC or just use GRAM. First step is to submit to a batch queue Next step is to see the optimization problem TeraGrid logistics will run our own MDS What does this mean to others batch queue prediction Probability number At that time, by that time are two different qns what needs to be supported in the language Add/remove nodes scheduler,X\ ,X  \ ,H yNeed a resource for 1 hr for time t Hard to do unless you use advanced reservations Need a resource for 1 hr by time t Can we have two separate language constructs? We propose we should Are we going to have slots for SC demo Looks like we need it to be able to support deadline Yang-Suk says we will have fixed slot Need to fix language a little bit Some stuff to support slot$0Q'[=$0Q'[  =[ QLong term question: How will vgES support application specific activities? Will callback that vgES have help? Pegasus represents data as tasks in workflow This will not work for LEAD because it has streaming data To play some interesting games we need some state object that can be used to reserve slots Not immediately but we need thisKZ#Z-Z;Z[Z!ZK#-;[! ,=Dynamic aspects of LEADSome slides (in Dennis stack from Houston meeting) (document on dynamic workflows) Motivation for nested workflows, control weather events control radars Dynamic workflows (still WIP in LEAD) Conditional and loops in workflows (next task is unknown) Can have weights for the possibilities This with constraints will be very tricky and interesting What do u do in the case where someone is not happy with your response and is willing to negotiate How does someone accept drift?  ZZ&ZZcZ Z &c   HReal-time prediction case need the schedule needs to be automated right now user driven How do we make sure we don t promise the same resources to more than one person? In general when failures happen, what do we do? Not making the constraint is just one of it Could be MTTF of resource? `BIBI /     0 0L(  Lr L S (  `  H L 0޽h ? 3380___PPT10.cPvr$mԈm1( / 0LDArialЃ 0Ѓ0hh0 0@ .  @n?" dd@  @@`` P>      0AA@3ʚ;ʚ;g4?d?d= 04ppp@ <4dddd4w 0h0 80___PPT10  LEAD-VGrADSDay 2 AMDiscussion TopicsvgES Performance Model Distributed control algorithms and how to handle workflow engines Implementation details of middle layer}vgESRefer to Yang-Suk s slides Nested slot  Can specify for a cluster or tight bag and applies to all children  hierarchical What needs to be specified to scheduler Hostname/<scheduler> allocationId NoOfProcessors Input/Output CPU time, wall time WS-GRAM RSL example How will vgES handle GT security? As far as possible keep the LEAD GRAM and vgLaunch similar BPSPtPStb A@   >Are we combining allocations? Options We have to do pre-submissions to the queue Optimization to reuse one allocations We manage our local pool with our own job manager Looks to PBS as one queue DVC needs to be started on every single node Globus gatekeeper on every single compute node? Seems like a lot of overheadj'ZQZ2ZxZZ'Q2x   nWe can submit job to DVC or just use GRAM. First step is to submit to a batch queue Next step is to see the optimization problem TeraGrid logistics will run our own MDS What does this mean to others batch queue prediction Probability number At that time, by that time are two different qns what needs to be supported in the language Add/remove nodes scheduler,X\ ,X  \ ,H yNeed a resource for 1 hr for time t Hard to do unless you use advanced reservations Need a resource for 1 hr by time t Can we have two separate language constructs? We propose we should Are we going to have slots for SC demo Looks like we need it to be able to support deadline Yang-Suk says we will have fixed slot Need to fix language a little bit Some stuff to support slot$0Q'[=$0Q'[  =[ QLong term question: How will vgES support application specific activities? Will callback that vgES have help? Pegasus   !"#$%&'C*+,-./0123456789:;<=>?VBDEFGHIJKLMNOPQ)TU(WXYZ[\]^_`abcdefghijklmnopqrstuvwxyz|}~Root EntrydO)pƽcS@Current User)SummaryInformation(XPowerPoint Document(DocumentSummaryInformation8H 0Ѓ0hh0 0@ .  @n?" dd@  @@`` P>      0AA@3ʚ;ʚ;g4?d?d= 04ppp@ <4dddd4w 0h0 80___PPT10  LEAD-VGrADSDay 2 AMDiscussion TopicsvgES Performance Model Distributed control algorithms and how to handle workflow engines Implementation details of middle layer}vgESRefer to Yang-Suk s slides Nested slot  Can specify for a cluster or tight bag and applies to all children  hierarchical What needs to be specified to scheduler Hostname/<scheduler> allocationId NoOfProcessors Input/Output CPU time, wall time WS-GRAM RSL example How will vgES handle GT security? As far as possible keep the LEAD GRAM and vgLaunch similar BPSPtPStb A@   >Are we combining allocations? Options We have to do pre-submissions to the queue Optimization to reuse one allocations We manage our local pool with our own job manager Looks to PBS as one queue DVC needs to be started on every single node Globus gatekeeper on every single compute node? Seems like a lot of overheadj'ZQZ2ZxZZ'Q2x   nWe can submit job to DVC or just use GRAM. First step is to submit to a batch queue Next step is to see the optimization problem TeraGrid logistics will run our own MDS What does this mean to others batch queue prediction Probability number At that time, by that time are two different qns what needs to be supported in the language Add/remove nodes scheduler,X\ ,X  \ ,H yNeed a resource for 1 hr for time t Hard to do unless you use advanced reservations Need a resource for 1 hr by time t Can we have two separate language constructs? We propose we should Are we going to have slots for SC demo Looks like we need it to be able to support deadline Yang-Suk says we will have fixed slot Need to fix language a little bit Some stuff to support slot$0Q'[=$0Q'[  =[ QLong term question: How will vgES support application specific activities? Will callback that vgES have help? Pegasus represents data as tasks in workflow This will not work for LEAD because it has streaming data To play some interesting games we need some state object that can be used to reserve slots Not immediately but we need thisKZ#Z-Z;Z[Z!ZK#-;[! ,=Dynamic aspects of LEADSome slides (in Dennis stack from Houston meeting) (document on dynamic workflows) Motivation for nested workflows, control weather events control radars Dynamic workflows (still WIP in LEAD) Conditional and loops in workflows (next task is unknown) Can have weights for the possibilities This with constraints will be very tricky and interesting What do u do in the case where someone is not happy with your response and is willing to negotiate How does someone accept drift?  ZZ&ZZcZ Z &c   DReal-time prediction case need the schedule needs to be automated right now user driven How do we make sure we don t promise the same resources to more than one person? In general when failures happen, what do we do? Not making the constraint is just one of it Could be MTTF of resource `BGBG /     0 0L(  Lr L S (  `  H L 0޽h ? 3380___PPT10.cPvrO6kfOl1( / 0LDArialЃ 0Ѓ0hh0 0@ .   Oh+'0( `h|   LEAD-VGrADSUNCUNC78Microsoft Office PowerPoint@0|=@רD*c@cG0g    y--$xx--'@Arial-.  2 6/LEAD."SystemH-@Arial-.  2 6I-.-@Arial-. 2 6LVGrADS.-@Arial-. 2 KADay 2 AM.-՜.+,0    On-screen ShowUNCK  G  ArialDefault Design LEAD-VGrADSDiscussion represents data as tasks in workflow This will not work for LEAD because it has streaming data To play some interesting games we need some state object that can be used to reserve slots Not immediately but we need thisKZ#Z-Z;Z[Z!ZK#-;[! ,=Dynamic aspects of LEADSome slides (in Dennis stack from Houston meeting) (document on dynamic workflows) Motivation for nested workflows, control weather events control radars Dynamic workflows (still WIP in LEAD) Conditional and loops in workflows (next task is unknown) Can have weights for the possibilities This with constraints will be very tricky and interesting What do u do in the case where someone is not happy with your response and is willing to negotiate How does someone accept drift?  ZZ&ZZcZ Z &c   LReal-time prediction case needs the VGrADS scheduler to be automated right now user driven How do we make sure we don t promise the same resources to more than one person? In general when failures happen, what do we do? Not making the constraint is just one of it Could be MTTF of resource? LEIEI,$ /     0 0L(  Lr L S (  `  H L 0޽h ? 3380___PPT10.cPvrŠt&1( / 0LDArialЃ 0Ѓ0hh0 0@ .  @n?" dd@  @@`` P>      0AA@3ʚ;ʚ;g4?d?d= 04ppp@ <4dddd4w 0h0 80___PPT10   LEAD-VGrADSDay 2 AMDiscussion TopicsvgES Performance Model Distributed control algorithms and how to handle workflow engines Implementation details of middle layer}vgESRefer to Yang-Suk s slides Nested slot  Can specify for a cluster or tight bag and applies to all children  hierarchical What needs to be specified to scheduler Hostname/<scheduler> allocationId NoOfProcessors Input/Output CPU time, wall time WS-GRAM RSL example How will vgES handle GT security? As far as possible keep the LEAD GRAM and vgLaunch similar BPSPtPStb A@   >Are we combining allocations? Options We have to do pre-submissions to the queue Optimization to reuse one allocations We manage our local pool with our own job manager Looks to PBS as one queue DVC needs to be started on every single node Globus gatekeeper on every single compute node? Seems like a lot of overheadj'ZQZ2ZxZZ'Q2x   nWe can submit job to DVC or just use GRAM. First step is to submit to a batch queue Next step is to see the optimization problem TeraGrid logistics will run our own MDS What does this mean to others batch queue prediction Probability number At that time, by that time are two different qns what needs to be supported in the language Add/remove nodes scheduler,X\ ,X  \ ,H yNeed a resource for 1 hr for time t Hard to do unless you use advanced reservations Need a resource for 1 hr by time t Can we have two separate language constructs? We propose we should Are we going to have slots for SC demo Looks like we need it to be able to support deadline Yang-Suk says we will have fixed slot Need to fix language a little bit Some stuff to support slot$0Q'[=$0Q'[  =[ QLong term question: How will vgES support application specific activities? Will callback that vgES have help? Pegasus represents data as tasks in workflow This will not work for LEAD because it has streaming data To play some interesting games we need some state object that can be used to reserve slots Not immediately but we need thisKZ#Z-Z;Z[Z!ZK#-;[! ,=Dynamic aspects of LEADSome slides (in Dennis stack from Houston meeting) (document on dynamic workflows) Motivation for nested workflows, control weather events control radars Dynamic workflows (still WIP in LEAD) Conditional and loops in workflows (next task is unknown) Can have weights for the possibilities This with constraints will be very tricky and interesting What do u do in the case where someone is not happy with your response and is willing to negotiate How does someone accept drift?  ZZ&ZZcZ Z &c   Real-time prediction case needs the VGrADS scheduler to be automated right now user driven How do we make sure we don t promise the same resources to more than one person? In general when failures happen, what do we do? Not making the constraint is just one of it Could be MTTF of resource? Reliability specification TEZZZdZEd$$ /     0 0L(  Lr L S (  `  H L 0޽h ? 3380___PPT10.cPvrbJ>12( / 0LDArialЃ 0Ѓ0hh0 0@ .  @n?" dd@  @@`` P>      0AA@3ʚ;ʚ;g4?d?d= 04ppp@ <4dddd4w 0h0 80___PPT10 [ LEAD-VGrADSDay 2 AMDiscussion TopicsvgES Performance Model Distributed control algorithms and how to handle workflow engines Implementation details of middle layer}vgESRefer to Yang-Suk s slides Nested slot  Can specify for a cluster or tight bag and applies to all children  hierarchical What needs to be specified to scheduler Hostname/<scheduler> allocationId NoOfProcessors Input/Output CPU time, wall time WS-GRAM RSL example How will vgES handle GT security? As far as possible keep the LEAD GRAM and vgLaunch similar BPSPtPStb A@   >Are we combining allocations? Options We have to do pre-submissions to the queue Optimization to reuse one allocations We manage our local pool with our own job manager Looks to PBS as one queue DVC needs to be started on every single node Globus gatekeeper on every single compute node? Seems like a lot of overheadj'ZQZ2ZxZZ'Q2x   nWe can submit job to DVC or just use GRAM. First step is to submit to a batch queue Next step is to see the optimization problem TeraGrid logistics will run our own MDS What does this mean to others batch queue prediction Probability number At that time, by that time are two different qns what needs to be supported in the language Add/remove nodes scheduler,X\ ,X  \ ,H yNeed a resource for 1 hr for time t Hard to do unless you use advanced reservations Need a resource for 1 hr by time t Can we have two separate language constructs? We propose we should Are we going to have slots for SC demo Looks like we need it to be able to support deadline Yang-Suk says we will have fixed slot Need to fix language a little bit Some stuff to support slot$0Q'[=$0Q'[  =[ QLong term question: How will vgES support application specific activities? Will callback that vgES have help? Pegasus represents data as tasks in workflow This will not work for LEAD because it has streaming data To play some interesting games we need some state object that can be used to reserve slots Not immediately but we need thisKZ#Z-Z;Z[Z!ZK#-;[! ,=Dynamic aspects of LEADSome slides (in Dennis stack from Houston meeting) (document on dynamic workflows) Motivation for nested workflows, control weather events control radars Dynamic workflows (still WIP in LEAD) Conditional and loops in workflows (next task is unknown) Can have weights for the possibilities This with constraints will be very tricky and interesting What do u do in the case where someone is not happy with your response and is willing to negotiate How does someone accept drift?  ZZ&ZZcZ Z &c   Real-time prediction case needs the VGrADS scheduler to be automated right now user driven How do we make sure we don t promise the same resources to more than one person? In general when failures happen, what do we do? Not making the constraint is just one of it Could be MTTF of resource? Reliability specification in vgES, diagnosis and recovery.TEZZZZE6$ /     0 0L(  Lr L S (  `  H L 0޽h ? 3380___PPT10.cPvr8r$12( / 0LDArialЃ 0Ѓ0hh0 0@ .  @n?" dd@  @@`` P>      0AA@3ʚ;ʚ;g4?d?d= 04ppp@ <4dddd4w 0h0 80___PPT10 [ LEAD-VGrADSDay 2 AMDiscussion TopicsvgES Performance Model Distributed control algorithms and how to handle workflow engines Implementation details of middle layer}vgESRefer to Yang-Suk s slides Nested slot  Can specify for a cluster or tight bag and applies to all children  hierarchical What needs to be specified to scheduler Hostname/<scheduler> allocationId NoOfProcessors Input/Output CPU time, wall time WS-GRAM RSL example How will vgES handle GT security? As far as possible keep the LEAD GRAM and vgLaunch similar BPSPtPStb A@   >Are we combining allocations? Options We have to do pre-submissions to the queue Optimization to reuse one allocations We manage our local pool with our own job manager Looks to PBS as one queue DVC needs to be started on every single node Globus gatekeeper on every single compute node? Seems like a lot of overheadj'ZQZ2ZxZZ'Q2x   nWe can submit job to DVC or just use GRAM. First step is to submit to a batch queue Next step is to see the optimization problem TeraGrid logistics will run our own MDS What does this mean to others batch queue prediction Probability number At that time, by that time are two different qns what needs to be supported in the language Add/remove nodes scheduler,X\ ,X  \ ,H yNeed a resource for 1 hr for time t Hard to do unless you use advanced reservations Need a resource for 1 hr by time t Can we have two separate language constructs? We propose we should Are we going to have slots for SC demo Looks like we need it to be able to support deadline Yang-Suk says we will have fixed slot Need to fix language a little bit Some stuff to support slot$0Q'[=$0Q'[  =[ QLong term question: How will vgES support application specific activities? Will callback that vgES have help? Pegasus represents data as tasks in workflow This will not work for LEAD because it has streaming data To play some interesting games we need some state object that can be used to reserve slots Not immediately but we need thisKZ#Z-Z;Z[Z!ZK#-;[! ,=Dynamic aspects of LEADSome slides (in Dennis stack from Houston meeting) (document on dynamic workflows) Motivation for nested workflows, control weather events control radars Dynamic workflows (still WIP in LEAD) Conditional and loops in workflows (next task is unknown) Can have weights for the possibilities This with constraints will be very tricky and interesting What do u do in the case where someone is not happy with your response and is willing to negotiate How does someone accept drift?  ZZ&ZZcZ Z &c   Real-time prediction case needs the VGrADS scheduler to be automated right now user driven How do we make sure we don t promise the same resources to more than one person? In general when failures happen, what do we do? Not making the constraint is just one of it Could be MTTF of resource? Reliability specification in vgES, diagnosis and recovery.TEZZZZE6$ /   r`<10( / 0LDArialЃ 0Ѓ0hh0 0@ .  @n?" dd@  @@`` P>      0AA@3ʚ;ʚ;g4?d?d= 04ppp@ <4dddd4w 0h0 80___PPT10 Y LEAD-VGrADSDay 2 AMDiscussion TopicsvgES Performance Model Distributed control algorithms and how to handle workflow engines Implementation details of middle layer}vgESRefer to Yang-Suk s slides Nested slot  Can specify for a cluster or tight bag and applies to all children  hierarchical What needs to be specified to scheduler Hostname/<scheduler> allocationId NoOfProcessors Input/Output CPU time, wall time WS-GRAM RSL example How will vgES handle GT security? As far as possible keep the LEAD GRAM and vgLaunch similar BPSPtPStb A@   >Are we combining allocations? Options We have to do pre-submissions to the queue Optimization to reuse one allocations We manage our local pool with our own job manager Looks to PBS as one queue DVC needs to be started on every single node Globus gatekeeper on every single compute node? Seems like a lot of overheadj'ZQZ2ZxZZ'Q2x   nWe can submit job to DVC or just use GRAM. First step is to submit to a batch queue Next step is to see the optimization problem TeraGrid logistics will run our own MDS What does this mean to others batch queue prediction Probability number At that time, by that time are two different qns what needs to be supported in the language Add/remove nodes scheduler,X\ ,X  \ ,H yNeed a resource for 1 hr for time t Hard to do unless you use advanced reservations Need a resource for 1 hr by time t Can we have two separate language constructs? We propose we should Are we going to have slots for SC demo Looks like we need it to be able to support deadline Yang-Suk says we will have fixed slot Need to fix language a little bit Some stuff to support slot$0Q'[=$0Q'[  =[ QLong term question: How will vgES support application specific activities? Will callback that vgES have help? Pegasus represents data as tasks in workflow This will not work for LEAD because it has streaming data To play some interesting games we need some state object that can be used to reserve slots Not immediately but we need thisKZ#Z-Z;Z[Z!ZK#-;[! ,=Dynamic aspects of LEADSome slides (in Dennis stack from Houston meeting) (document on dynamic workflows) Motivation for nested workflows, control weather events control radars Dynamic workflows (still WIP in LEAD) Conditional and loops in workflows (next task is unknown) Can have weights for the possibilities This with constraints will be very tricky and interesting What do u do in the case where someone is not happy with your response and is willing to negotiate How does someone accept drift?  ZZ&ZZcZ Z &c   Real-time prediction case needs the VGrADS scheduler to be automated right now user driven How do we make sure we don t promise the same resources to more than one person? In general when failures happen, what do we do? Not making the constraint is just one of it Could be MTTF of resource? Reliability specification in vgES, diagnosis and recoveryTEZZZZE6$ /     0 0L(  Lr L S (  `  H L 0޽h ? 3380___PPT10.cPvr1