Charm++ parallel programming system

[Balaji S <balaji.ceg.13 AT gmail.com>]

Sir, 
  Thanks for the useful info. I have two doubts,
1. I get bgTrace* files when i run with +bglog option in the files in charm/examples/bigsim/example. However for normal charm++ programs, how can i get the bgTrace*

files, like the files located in charm/examples/charm++ ? 
2. I used the projections tool as u said.. It takes sts files as input. But wat purpose does the log files pgm.0.log ,... pgm.3.log serve? Could we intrepret anything from them?

Similarly log2proj tool produces only log files not sts files so how can we run them in projections tool?

Please help sir.

On Fri, Mar 2, 2012 at 2:13 AM, Mokos, Ryan <mokos AT illinois.edu> wrote:

BigNetSim doesn't create output files.  However, you can view network link statistics by changing DISPLAY_LINK_STATS in the netconfig file to 1.

Ryan

---

From: Balaji S [balaji.ceg.13 AT gmail.com]
Sent: Thursday, March 01, 2012 1:31 PM

To: Mokos, Ryan
Subject: Re: [ppl] [charm] Error when running BigNetSim

Sir, I just gave Make in the BigNetSim/trunk.  Ya Biggene works properly. I have attached the output
when i ran,
 ./charmrun +p2 ./bigsimulator 0 1
on the bgTrace* in the SampleTraces directory. Wont i get any output files? analyzing the trafffic patterns etc.,

On Fri, Mar 2, 2012 at 12:31 AM, Mokos, Ryan <mokos AT illinois.edu> wrote:

I would ask the charm group.  I'm not familiar with the characteristics of most of the tests and examples.

Ryan

---

From: Balaji S [balaji.ceg.13 AT gmail.com]
Sent: Thursday, March 01, 2012 12:53 PM

To: Mokos, Ryan
Subject: Re: [ppl] [charm] Error when running BigNetSim

Sir, and also, where can i get charm++ benchmarks , say with 100's of messages passed ( just to simulate and see, though i am a noob :) )

On Fri, Mar 2, 2012 at 12:13 AM, Balaji S <balaji.ceg.13 AT gmail.com> wrote:

Thanks for the useful info sir. I am getting this error at the end , when i typed 'make'
make[1]: Leaving directory `/home/balaji/newcharm/charm/BigNetSim/trunk/KaryNmesh'
make[1]: *** make[1]: Entering directory `/home/balaji/newcharm/charm/BigNetSim/trunk/KaryNmesh'
No rule to make target `test'.  Stop.
make[1]: Leaving directory `/home/balaji/newcharm/charm/BigNetSim/trunk/KaryNmesh'
make: *** [test] Error 1

i have attached the terminal output when the 'make' runs.
Thank you sir.

On Thu, Mar 1, 2012 at 10:58 PM, Mokos, Ryan <mokos AT illinois.edu> wrote:

Hi Balaji,

What errors are you getting?  I thought you were able to compile BigNetSim a couple days ago.  Did you do something different?

This is the basic format:

[A] B name:C (srcpe:D msgID:E) ep:F charm_ep:G
 recvtime:H startTime:I endTime:J execTime:K
-msgID:L sent:M recvtime:N dstNode:O tid:P size:Q group:R
backward: S
forward: T

A - event number
B - address that you don't need to worry about
C - event name (= msgep if the event is a message receive)
D - source PE for a message receive; = -1 if it's not from receiving a message
E - message ID for a message receive; = -1 if it's not from receiving a message
F - entry point/entry method number (don't worry about this)
G - charmentry point/entry method number (don't worry about this)
H - time (in seconds) at which the event is received if it's a message; = -1 if event is not a message receive
I - time (in seconds) at which the event begins execution
J - time (in seconds) at which the event ends execution
K - event execution time (in seconds)

If messages are sent, then you will see "-msgID: ..." lines.
L - message ID
M - time (in seconds) at which message was sent
N - time (in seconds) at which message was received by destination PE
O - destination node - can be a single node or multiple nodes (e.g., broadcast all)
P - destination thread ID - basically the number of the destination core on the destination node
Q - message size in bytes
R - destination node grouping - set to 1 for point-to-point messages and broadcasts; set > 1 for multicasts with subsequent messages that are part of the multicast set to -1

S - list of backward dependent event numbers (don't worry about the accompanying addresses)
T - list of forward dependent event numbers (don't worry about the accompanying addresses)

Note that forward and backward dependencies only point to events in that time line.  Messages can point to events in the same time line or other time lines.

Compile text2log and run with -help to get some more information, especially on message destination node codes.

You can also extract the examples tarball and look through all the cases.  Go through them in order, starting with case0, as they grow in complexity.  Only look at the detail*.log files--these are the ASCII dumps of the bgTrace files (the log*.txt files are the input files that text2log used to create the bgTrace files; they are similar to ASCII dumps, but don't contain all the info as text2log fills some of it in automatically).  The file name states the PE (i.e., time line) number: detail-0-0-0.<PE>.log.  I'm attaching a file to this e-mail that gives a graphical representation of each case.

Ryan

---

From: Balaji S [balaji.ceg.13 AT gmail.com]
Sent: Thursday, March 01, 2012 7:20 AM

To: Mokos, Ryan
Subject: Re: [ppl] [charm] Error when running BigNetSim

Hi Sir,
 Thanks for the reply. I ll try out the various options. But i get errors while i 'make' BigNetSim. I checked out latest Bignetsim and
downloaded Charm from http://charm.cs.illinois.edu/autobuild/cur/ Is it the right version?

And also one more doubt sir, ( Sorry for troubling you) I do not know how to intrepret the log files, ( when i convert bgtrace* to ASCII in loganalyzer ) Any help sir? how to find  the src node and destination nodes of the messages?
 

On Thu, Mar 1, 2012 at 12:29 AM, Mokos, Ryan <mokos AT illinois.edu> wrote:

Hi Balaji,

Thanks for sending your trace files--they definitely are not valid.  Ring generated 4 separate time lines.  Two of them don't have any events in them (you can see this by running the LogAnalyzer tool in the same directory as the bgTrace files), and the other two each have one strange event (e.g., one of them is the receipt of a message from PE 6, which doesn't exist).

I tried running ring the way you did.  One problem is that it appears to create a number of worker threads equal to x*y*z*<number of specified worker threads>.  So when you specify 2 workers, it generates 2x the number of time lines, and all the extras have no events in them.  Furthermore, even when I specify 1 worker, I still get bizarre time lines.  For example:

======================== Run ===========================

./ring 2 1 1 1 1 +bglog
Charm++: standalone mode (not using charmrun)
Converse/Charm++ Commit ID: v6.3.0-1228-g34cdf5e
Charm++> scheduler running in netpoll mode.
BG info> Simulating 2x1x1 nodes with 1 comm + 1 work threads each.
BG info> Network type: bluegene.
alpha: 1.000000e-07     packetsize: 1024        CYCLE_TIME_FACTOR:1.000000e-03.
CYCLES_PER_HOP: 5       CYCLES_PER_CORNER: 75.
BG info> cpufactor is 1.000000.
BG info> floating point factor is 0.000000.
BG info> Using WallTimer for timing method.
BG info> Generating timing log.
0 0 0 => 1 0 0
1 0 0 => 0 0 0
0 0 0 => 1 0 0
[0] Number is numX:2 numY:1 numZ:1 numCth:1 numWth:1 numEmulatingPes:1 totalWorkerProcs:2 bglog_ver:6
[0] Wrote to disk for 2 BG nodes.

BG> BigSim emulator shutdown gracefully!
BG> Emulation took 0.006090 seconds!
Program finished.

======================== Time Line 0 ===========================

[0] 0x618070 name:msgep (srcpe:1 msgID:1) ep:2 charm_ep:-1
 recvtime:0.000010 startTime:0.000010 endTime:-1.000000 execTime:0.000000
backward:
forward:

======================== Time Line 1 ===========================

[0] 0x618070 name:msgep (srcpe:3 msgID:0) ep:2 charm_ep:-1
 recvtime:0.000000 startTime:0.000000 endTime:0.000010 execTime:0.000010
-msgID:1 sent:0.000010 recvtime:0.000010 dstNode:0 tid:-1 size:60 group:1
backward:
forward:

It looks like the simulation is starting with time line 1 (i.e., PE 1), which shouldn't happen.  I'm copying the rest of the development group so someone who knows about this code can take a look at this and see what's wrong.  In any case, I wouldn't use the traces generated by this code right now.

If you just want some trace files to run, you can find some that work at BigNetSim/trunk/sampleTrace.  Alternatively, you could do an AMPI build of charm with the bigemulator target and run one of several examples in charm/net-linux-x86_64/examples/ampi.  The build line (if you're using 64-bit Linux) should look something like:

./build AMPI net-linux-x86_64 bigemulator -j8 -g

Ryan

---

From: Balaji S [balaji.ceg.13 AT gmail.com]
Sent: Friday, February 24, 2012 9:50 PM

To: Mokos, Ryan
Subject: Re: [ppl] [charm] Error when running BigNetSim

HI sir,
     i have attached the trace files. I got those when i ran,
 ./ring 2 1 1 2 2  +bglog

in the charm/examples/bigsim/emulator directory.
Kindly help sir.

On Fri, Feb 24, 2012 at 2:33 PM, Mokos, Ryan <mokos AT illinois.edu> wrote:

Hi Balaji,

For some reason, it looks like it wasn't able to load the traces properly.  Please send me your bgTrace files so I can see what they contain.

Ryan

---

--
With Regards,
    Balaji.S

--
With Regards,
    Balaji.S

--
With Regards,
    Balaji.S

--
With Regards,
    Balaji.S

-- 
With Regards,
    Balaji.S