Table of Contents
Advertisement
FC uses link level buffer-to-buffer credit buffering to regulate data traffic. Prior to data
transmission, the receiver sends a number of credits to the transmitter (or the transmitter may
start with a default value). Each credit typically represents one FC frame of data. The
transmitter, upon receipt of the credits, has the authority to send data at any time up to the
equivalent number of credits that have arrived from the receiver.
Assuming it is not congested, the receiver immediately replenishes the credits after receiving
data from the transmitter. A well designed system will have sufficient credits to compensate
for the latency between the nodes. Thus, the transmitter will have enough credits in store to
allow the receiver to get the data and then send new credits back before the transmitter runs out
of credits. In this way, data transmission continues at the full rate of the FC link.
Should the final receiving unit become congested, it can exert back pressure against the FC
port simply by not re-supplying credits. The gateway, sensing a lack of credits from the FC
port, creates back pressure against the SONET interface so that the remote gateway can slow
its transmission.
The total round trip transmission time - the latency - in the system is a critical variable in flow
control. This latency has several components, the largest of which is the 5 µsec per km latency
between the two installations (the speed of light in optical fiber).
The LightSand S-2500B Gateway, designed specifically for Fibre Channel over SONET
transport, avoids the unnecessary latency that results from TCP retransmission.
In the critical area of buffer design, the S-2500B Gateway buffers are sized to store an amount
of data equivalent to the need created if full latency is experienced over a full-rate link. Full
rate transmission for the S-2500B can be sustained for up to 2000 km.
IP Over SONET
The consolidation of data in a centrally located domain, data-center to data-center backup, and
similar applications all employ clustered servers both at the local and the remote sites. In
addition to providing backup, a complete SAN extension for these applications is expected to
provide full availability to the remote data through a transparent extension of the LAN.
The same limitations in reliability in the use of conventional IP transport for extending FC also
apply to extending LAN communication. Highly available systems in remote sites require
clustered servers. Clustered servers in turn require a tightly coupled heartbeat signal indicating
operation. This heartbeat signal should not be processed through IP routers, which will drop
data in response to congestion.
Introduction
1-3
Advertisement
Table of Contents
