codes simplenet

Simplenet Network Model

Model overview:

Simplenet is a module that implements a network module for use as a component within model net.

The simplenet model is a basic cost model based on the startup cost (latency) of the network and the bandwidth (cost per byte transferred) of the network. Each NIC is assumed to have full-duplex transmit and receive capability. Messages begin transmission when the sender's output queue is available. From the sender's perspective, the transmission is complete as soon as enough virtual time has elapsed to account for the startup and bandwidth costs of the message. However, the message is not delivered for processing on the receiver until the receiver's input queue is also available. The network fabric itself is not modeled and is instead treated as if it has infinite buffering ability and no internal routing overhead.

Simplenet supports optimistic mode and reverse computation.

Modularization concepts:

The simplenet LPs are intended to be used in conjunction with other LPs that implement a higher level protocol. For examples, they may be used in conjunction with LPs that model file servers in a parallel file system.

Each simplenet LP represents a NIC. In terms of LP mapping, each simplenet LP should also use an LP ID that is a fixed offset from the "node" that it belongs to. Note that the offset is specified to the simplenet module at event creation time, which means that multiple "node" LPs can share the same simplenet LP by using different offsets.

In terms of event transmission, simplenet can be thought of as a wrapper around other events. It tunnels the upper level events through the network and delivers them to a remote LP, consuming some amount of virtual time along the way according to the modeled network characteristics. The upper level model does not have access to the simplenet LPs, the simplenet message struct, or any other internal parameters. It is intended to abstract away network details from some higher level model.

Simple P2P Network model

Model overview:

SimpleP2P is an extension of simplenet to allow for arbitrary point-to-point capacities.

SimpleP2P has nearly the same duplex queued transmission semantics as simplenet, though with a few important differences. First, point-to-point latencies and bandwidths are different. Network links can also have different ingress and egress latencies/bandwidths. Second, each unique link is given its own queue that do not interfere with the others. The use case for this model is a set of sites with internal networks, with each site communicating via simplep2p; we assume the internal network is capable of servicing all outgoing/incoming WAN transmissions at their full capacities.

Additional configuration is needed to initialize the link latencies/capacities. In the codes-configuration file, the variables "net_latency_ns_file" and "net_bw_mbps_file" must be set under the PARAMS group. They point (in a path relative to the configuration file) to configurations for the latency and bandwidth costs, respectively.

Each of the latency/bandwidth configuration files have the same format, based on a square matrix of directed point to point capacities. Given N modelnet LPs, it has the format:

1:1 1:2 ... 1:N
2:1 2:2 ... 2:N
      ...
N:1 N:2 ... N:N

where x:y is a pair of latency or bandwidth between components x and y. An individual entry in this matrix is specified as "a,b" where 'a' is the egress latency/bandwidth of the link and 'b' is the ingress latency/bandwidth of the link. Whitespace is ignored, but linebreaks are not, and delimit rows of the matrix. The relative simplep2p identifiers 1..N are assigned to simplep2p LPs in the order of their appearance in the codes-configuration file. It is expected that all i:i entries are 0 - modelnet currently doesn't handle self messages.

Support in the code is also available for triangular matrices of the format:

1:2 1:3   ...   1:N
2:3 2:4 ...     2:N
        ...
                N-1:N

However, this option is currently disabled (the configuration code path has not been expanded to allow specifying the option). The option will be enabled into the configuration routines if deemed necessary.

Caveats:

The model-net statistics are slightly more complex than in other model-net implementations because there are in essence multiple queues per simplep2p instance. In particular, the "send time" and "recv time" stats are computed as ranges of time in which the simplep2p LP is actively sending/receiving data from any link. Hence, simple bandwidth calculations (send bytes / send time) may not be accurate due to skew in message issuance.

Having more than one category in simplep2p will cause the times in the derived "all" category to be off. "all" is currently implemented as the sum of the various categories, which doesn't work when times in certain categories may overlap.