Curtiss-Wright GT200, Справочник по аппаратуре, Страница: 32 / 62

OPERATION
4.2.10 Latency
Latency is the delay that a frame accumulates when passing through nodes and cables.
Each node contributes approximately 0.45 µs to a frame’s latency, and each meter of
fiber-optic cable contributes 5 ns to a frame’s latency. A complete discussion of GT200
network latency is beyond the scope of this manual. For more information, see
Curtiss-Wright Controls Technote G-T-ST-SCTN0138-A-0-A1, or consult with our Tech
Support staff.
4.2.11 Throughput
Sustained write throughputs of 205 MBps are possible using DMA on high performance
64 bit/66 Mhz platforms. Read throughput is not limited by network bandwidth.
4.2.12 Message Ordering
Host writes to the network progress in-order; message priorities are not used. Once on the
network, messages are not reordered. However, new messages from other nodes may be
inserted between messages that are in progress.
4.2.13 Memory
In its simplest form, the GT200 Network system appears as general-purpose memory.
The use of this memory depends on the nature of the data stored and the conventions and
limitations imposed by the specific host computer system and operating system. On most
processors, this means that the application program can use this memory in the same way
as any other data-storage area of memory. The memory cannot be used as instruction
space.
The major difference between GT200 memory and system memory is that any data
written into GT200 memory is automatically sent to the same GT200 memory location in
all nodes on the network. This is why it is also referred to as replicated shared memory.
When a host computer writes to the shared memory, the GT200 node host adapter
supplies the proper handshaking logic. The shared memory behaves somewhat like
resident or local memory.
A device driver is required for operating systems. The device driver performs the
following functions: configure the GT200, maps memory, handles interrupts, transfers
user buffers to/from network memory.
When performing a transfer of user buffer to/from network memory, the driver works in
conjunction with the operating system and the GT200, and may break the user buffer into
multiple DMA transfers. This is a result of paging, physical memory requirements, and
the need to have a contiguous memory region for the DMA descriptors.
4.2.14 FIFO Buffers
The GT200 card contains various FIFO buffers used for temporarily storing information
during normal send and receive operation of the node.
4.2.14.1 Retransmit FIFO
This buffer is used to receive foreign messages from the network, and send them on, or to
hold received foreign messages while inserting a native message from the host onto the
network.
Copyright 2005 4-4 GT200 Hardware Reference