Re: [Gems-users] Node number limitations?

[Dan Gibson <degibson@xxxxxxxx>]

Marco/Enric,

We are able to use >64 (and >32) banks and/or processors on both 64-bit 
and 32-bit machines. Since you have the same issue with BigSet as with 
OptBigSet, I think we can rule out either of those files as the source 
of the problem.
Try using the "nq" debug flags to examine message routing and queuing. 
Its unlikely that there's a bug in the network generation code (as its 
been used extensively), though specifying topologies and/or protocols is 
often a bug-prone process.
Regards,
Dan

Enric Herrero wrote:
> Thank you,
>
> We asked Mike some time ago and he said the same. But using BigSet 
> gives us the same error. He suggested that it may be a bug in the 
> network generation code. However we don't know where is this code 
> generated.
> Sorry for asking this twice but we are a little bit lost in the ruby code.
>
> Regards,
>
> /Marco and Enric
>
> */Dan Gibson <degibson@xxxxxxxx>/* escribió:
>
>     The code you are asking about is intended to handle the case where a
>     "set" consists of multiple words, most of which will be all-1's,
>     but the
>     topmost will have some of its most-significant bits zeroed. That
>     is the
>     purpose of the mask associated with that comment.
>
>     OptBigSet in general is optimized for 64-bit machines, so the
>     simplest
>     way to find out if OptBigSet is your problem would be to use BigSet
>     instead. To do so, edit Set.h and #define BIGSET instead of #define
>     OPTBIGSET.
>
>     Regards,
>     Dan
>
>     Enric Herrero wrote:
>     > Hi, Thank you for answering.
>     >
>     > We use a 32 bit processor and despite setting the file to #define
>     > __32BITS__ is still not working.
>     >
>     > We have also problems with the number of memories used and seems
>     to be
>     > the same kind of problem. When we use 64 processors and we try to
>     > access node 55 it is going to node 23.
>     >
>     > Desired node 55 -> 110111
>     > Node accessed 23 -> 10111
>     > As we see ruby seems to use less bits than the necessary for
>     dealing
>     > with this number of memories. We have seen that in OptBigSet.C when
>     > the vectors are defined some masks are applied. Why are these masks
>     > applied?
>     > We don't understand the sentence:
>     >
>     > // the number of populated spaces in the higest-order array slot is:
>     > // m_nSize % 32, so the uppermost 32 - m_nSize%32 bits should be
>     > // cleared
>     >
>     > Is this related to our problem? And if it has relation it is
>     possible
>     > to increase the number of bits used?
>     >
>     > Thank you
>     >
>     > /Marco and Enric
>     >
>     >
>     > */Dan Gibson /* escribió:
>     >
>     > Marco/Enric,
>     > I think this is a known bug with the file OptBigSet.C, which
>     > implements bit vectors. With some compiler/platform combinations,
>     > the macros that OptBigSet.C uses are not properly set. The
>     > solution is relatively straightforward: Edit OptBigSet.C, and find
>     > the code that defines either __64BITS__ or __32BITS__, depending
>     > on your platform, and simply remove the #if condition that selects
>     > which marco is used. Thus, you might replace:
>     > #if __amd64__ || __LP64__
>     > #define __64BITS__
>     > #else
>     > #define __32BITS__
>     > #endif
>     > with:
>     > #define __32BITS__
>     >
>     > You should also add a #define __32BITS__ to OptBigSet.h.
>     >
>     > Regards,
>     > Dan Gibson
>     >
>     >
>     > Marco Tirado wrote:
>     >> Hello:
>     >>
>     >> We are simulating a CMP system with Ruby. We have succesfully
>     >> simulated the system with 32 processors. However after increasing
>     >> the number of nodes in the Network topology file (this for a
>     >> FILE_SPECIFIED network) we get the following error:
>     >>
>     >> failed assertion 'msg_destinations.count() == 0' at fn
>     >> virtual void PerfectSwitch::wakeup() in
>     >> network/simple/PerfectSwitch.C:264
>     >>
>     >> At first we were using 64 nodes in the network and it worked
>     >> fine, we get the error when we increase the number of nodes to
>     >> 96. Since this is the only parameter we have changed in the
>     >> system, is there any GEMS limitation for the number of nodes or
>     >> in the number of links ? Any help on this will be apreciated.
>     >>
>     >> Marco and Enric
>     >>
>     ------------------------------------------------------------------------
>     >>
>     >> _______________________________________________
>     >> Gems-users mailing list
>     >> Gems-users@xxxxxxxxxxx
>     >> https://lists.cs.wisc.edu/mailman/listinfo/gems-users
>     >>
>     > _______________________________________________
>     > Gems-users mailing list
>     > Gems-users@xxxxxxxxxxx
>     > https://lists.cs.wisc.edu/mailman/listinfo/gems-users
>     >
>     >
>     >
>     ------------------------------------------------------------------------
>     >
>     > LLama Gratis a cualquier PC del Mundo.
>     > Llamadas a fijos y móviles desde 1 céntimo por minuto.
>     > http://es.voice.yahoo.com
>     >
>     >
>     >
>     ------------------------------------------------------------------------
>     >
>     > _______________________________________________
>     > Gems-users mailing list
>     > Gems-users@xxxxxxxxxxx
>     > https://lists.cs.wisc.edu/mailman/listinfo/gems-users
>     >
>     _______________________________________________
>     Gems-users mailing list
>     Gems-users@xxxxxxxxxxx
>     https://lists.cs.wisc.edu/mailman/listinfo/gems-users
>
>
> ------------------------------------------------------------------------
>
> LLama Gratis a cualquier PC del Mundo.
> Llamadas a fijos y móviles desde 1 céntimo por minuto.
> http://es.voice.yahoo.com 
> <http://us.rd.yahoo.com/mail/es/tagline/messenger/*http://es.voice.yahoo.com/> 
>
> ------------------------------------------------------------------------
>
> _______________________________________________
> Gems-users mailing list
> Gems-users@xxxxxxxxxxx
> https://lists.cs.wisc.edu/mailman/listinfo/gems-users
>