Re: [Gems-users] Bit-level analysis of the on-chip messages

[Niket <niketa@xxxxxxxxxxxxx>]

> Actually I don't need the bit-level "values", but I want to know the 
> bit length of each field in a message. For example, a control message 
> possibly contains requester ID, destination ID, some flags, ect, and I 
> am wondering how to know the bit length of each of these fields. In 
> other words, I want to know the "format" of a message. As you 
> mentioned, it depends on the protocol. To my understanding, GEMS 
> treats a message as a "bulk" structure, i.e. it defines the contents 
> of a message, but transmits it as a whole thing (8 bytes or 64+8 
> bytes). Am I right?
GEMS does transfer the message in a 'bulk' at a time. If you want to 
know the bit length, you can look at the protocol and the system 
configuration and come up with your own bit lengths for your 
calculations. But as Mike said, GEMS doesn't care about the individual 
field lengths. It assumes that the control information can be fit into 
the 8 bytes and seems to be a fair assumption.
> How does the new network model deal with the messages? Do you have any 
> idea about this?
In the detailed network, although the messages will  be broken into 
network level units, I do not think the fields will be mapped 
separately. Actually the latency of packet delivery is not dependent on 
these fields.
> Thanks,
> Lide
>
> On 1/17/08, *Niket* <niketa@xxxxxxxxxxxxx 
> <mailto:niketa@xxxxxxxxxxxxx>> wrote:
>     Hi Lide,
>
>     The control messages are all 8 bytes but are dynamically casted
>     depending on the protocol. For instance, a request message might be a
>     token request in a token coherence protocol or an invalidation in a
>     directory protocol. You can figure that out  by  making minor
>     modifications.
>
>     The data messages are basically 64 byte cache lines + 8 byte control
>     information. If you are looking for data values, I am not sure
>     whether
>     Ruby carries the data values around. From what I understand, it is a
>     timing interface and hence cache line values do not fly around. I
>     remember reading some old posts mentioning that some previous version
>     had the interface but I am not sure it is any longer present.
>
>     But why do you need these bit-level values ? I presume you want to do
>     some power calculations ?
>
>     As regards the detailed network model inside GEMS, it should be out
>     pretty soon.
>
>     Regards,
>     Niket
>
>
>     Lide Duan wrote:
>     > Hi everyone,
>     >
>     > I am wondering if there is any bit-level configuration/analysis
>     of the
>     > messages traversing across the on-chip network in GEMS. I notice
>     that,
>     > for the various on-chip messages, GEMS specifies the control
>     message
>     > as 8 bytes and the data message as 64+8 bytes, but I need more
>     > detailed information about what these bytes/bits should be. I also
>     > notice that, for each protocol, there is a file that specifies the
>     > message structures used in this protocol, but it does not
>     contain the
>     > bit length information.
>     >
>     > Besides, I am also wondering when the next version GEMS that
>     > incorporates a detailed network model will be released. I am quite
>     > expecting it.
>     >
>     > Thanks,
>     > Lide
>     >
>     ------------------------------------------------------------------------
>     >
>     > _______________________________________________
>     > Gems-users mailing list
>     > Gems-users@xxxxxxxxxxx <mailto:Gems-users@xxxxxxxxxxx>
>     > https://lists.cs.wisc.edu/mailman/listinfo/gems-users
>     > Use Google to search the GEMS Users mailing list by adding
>     "site: https://lists.cs.wisc.edu/archive/gems-users/"; to your search.
>     >
>     >
>
>     _______________________________________________
>     Gems-users mailing list
>     Gems-users@xxxxxxxxxxx <mailto:Gems-users@xxxxxxxxxxx>
>     https://lists.cs.wisc.edu/mailman/listinfo/gems-users
>     Use Google to search the GEMS Users mailing list by adding "site:
>     https://lists.cs.wisc.edu/archive/gems-users/"; to your search.
>
>
> ------------------------------------------------------------------------
>
> _______________________________________________
> Gems-users mailing list
> Gems-users@xxxxxxxxxxx
> https://lists.cs.wisc.edu/mailman/listinfo/gems-users
> Use Google to search the GEMS Users mailing list by adding "site:https://lists.cs.wisc.edu/archive/gems-users/"; to your search.
>
>