Ptolemy EVM Simulation

1
Ptolemy EVM Simulation

• Current options
• -evm
• BM on Myrinet, RM uses simple external RCN
• Trigger
• Poisson (default), or Gaussian with cut-off
• Event fragments default LogNormal
• BU emulates filter farms
• Farm processing time
• LogNormal (default)
• Gaussian with cut-off
• Input histogram
• -evma
• -evm RCN over Myrinet
• -evmb
• -evma Barrel Shifter on Builder Unit side

2
RCN over Myrinet
3
Configuration for plots

• 63x63 system BM RM
• RCN over Myrinet (-evma)
• Trigger 12.5 kHz (Poisson)
• Message event ID packing RCN 2, BCN 1
• Event fragments LogNormal 16/-16 kB/RU, unless
  otherwise specified
  
• Farm processing time (FPT) LogNormal, values
  given per plot
  
• Each BU emulates 8 filter nodes
• Average FPT should not exceed 40 ms/event
• BU returns an event ID at the end of its FPT (!)

4
Poisson trigger distribution
ms
5
Event fragment distribution
6
Active events in EVM, FPT 40 /- 0ms
ms
7
Farm Processing Time Distribution, 40/-20 ms
ms
8
Active events in EVM, FPT 40/-20ms
Steady state level increase supposedly due to
queuing in filter nodes. Event with large FPT wil
l delay return of all following events sent to
that filter node. System should stabilize if FPT
average does not exceed 40 ms.
ms
9
Active events in EVM, FPT 40/-40ms
ms
10
Myrinet RCN Latency
ms
11
RCN latency with BS on BU side (-evmb)
ms
12
Active events in EVM, FPT 30/-30 ms
ms
13
RU max. memory usage (FPT 30/-30 ms)
As long as the BUs have free resources, this is
independent of the FPT (modulo differences in
random generator usage).
ms
14
BU max. active events, FPT 30/-30 ms
ms
15
BU max. memory usage, FPT 30/-30 ms
ms
16
Active events in EVM, FPT 30/-30 ms, Event
Fragments 16 /- 16/sqrt(8) kB/RU
ms
17
Active events in EVM, FPT 0/-0 ms,trigger
12.5/-0 kHz, event fragments 16/-0 kB
ms
18
Conclusions and outlook

• RCN over Myrinet appears to work
• Baseline?
• Farm emulation can use HLT event processing time
  input histograms
  
• Still need to resolve Ptolemy crashes at 8.5 s
  of running
  
• Calendar queue time-ordering problem
• Possibly due to floating point round-off
• What else needed for the EVM simulation?
• Compare with test stands what configurations?
• Error injection?