Fiber Channel Protocol (HDS White Paper)

I. Basic architecture of Fiber Channel
FC-4 upper layer protocol: SCSI, hippi, sbccs, 802.2, ATM, Vi, IP
The FC-3 common service
FC-2 (framing protocol/Flow Control)
FC-1 encode/decode
FC-0 media: optical or copper, 100 mb/sec to 1.062 GB/sec
Description:
FC-0: physical layer, custom different media, transmission distance, signal mechanism standard, also defines the optical fiber and copper interface and electrical
Cable Index
FC-1: defining encoding and decoding standards
FC-2: defines frame, stream control, and service quality
FC-3: defines common services such as data encryption and Compression
FC-4: Protocol ing layer, defines the interface between the Fiber Channel and the upper application, the upper application such as: Serial SCSI
Discussed, HbA drivers provide FC-4 interface functions, FC-4 support for multiple protocols such as: FCP-SCSI, FC-IP, FC-vi.
2. FCP-SCSI
FCP-SCSI: Is to map Fiber Channel devices into an operating system
Accessible
A serial protocol of the logical drive. This protocol allows the previous SCSI-based applications to use the fiber channel without any modifications. FC-SCSI is a storage system
And Server
The most important means of communication. SCSI extends the Copy command
, A New ANSI T10 standard that also supports San-based storage systems
Data can be directly moved through data migration applications.
The advantages of FCP-SCSI and bus connection in the storage LAN
The FCP-SCSI has been proven to provide higher performance (100 m/sec) with a further connection distance (up to 10 km per connection ), A larger addressing space (up to 16000000 nodes ). The FCP-SCSI replaces block transfer with frame transfer. Frame transmission transmits short and small transaction data in the transmission mode of large data streams, which improves the service quality. FCP-SCSI support for simplified management
And the storage "pool" Technology of resource storage
Network
Configuration. FCP-SCSI supports coding technology to improve reliability and availability
.
FC-IP
FC-IP maps fibre channel addresses to IP addresses, FC-IP addressing: broadcasts an IP address and returns a MAC address from the storage Node
Address. IP broadcast may cause errors if the SCSI device cannot differentiate between FCP-SCSI frames and FC-IP frames. HDS System can pass detection
Frame headers are used to distinguish between FCP-SCSI frames and FC-IP frames. storage systems without this capability must block FC-IP frames from being broadcast to a fiber port through other methods, such as switch zoning.
.
FC-IP has several advantages over Ethernet: It can be integrated with an internal connection architecture similar to FCP-SCSI storage to save costs; faster transmission and higher efficiency.
The maximum number of packets transmitted over Ethernet is 1500 bytes. The package is a basic unit in Ethernet. After each frame, the CPU cycle is interrupted. Load in GB Ethernet is usually a limiting factor to avoid occupying full bandwidth.
. While the FC-IP data frame reaches 2000 bytes, the basic unit of FC-IP correction is a multi-frame queue. MTU can reach 64 frames. Compared with Ethernet, fiber channel is allowed on the host.
More data is transmitted between interruptions. This MTU can reduce the CPU cycle and improve the transmission efficiency.
FC-IP also has the advantage of using fiber channel network, fiber channel network is based on flow control closed network. Ethernet is initially designed to pass through a public network without traffic control. When blocking occurs, it returns the resend packet after the consistent time period, consuming additional CPU cycles. The IP application runs on the FC-IP without modification, enjoying the high speed and greatly reducing processing interruptions brought by fiber channel.
Emulex and JNI are Fiber Channel HbA vendors that provide FC-IP drives. They plan to pass a "combo" to support FCP-SCSI and FC-IP. Troika provides controllers that support FCP-SCSI, FC-IP, and FC-IP (QOS), and QoS allows network administrators to assign protocol priority.
4. FC-VI
The FC-VI is implementing a VI architecture on fiber channels, which allows fast data migration between memory addresses at fiber channel points. FC-VI is the protocol standard for Vi-based fiber channel applications, an Intel, Compaq, more than 100 vendors and organizations in order to reduce server communication waits. VI was designed to achieve cluster computers
The effect of Inter-communication waiting for reduction and high bandwidth. In the fiber channel network, the target is achieved through direct access (DMA) between the HBA buffer of the interface with another node and the application memory.
The VI architecture establishes a memory registration mechanism, which is essentially to restrict users.
Memory Address and supports data from the user
The memory is directly transmitted to the HBA cache, and then the data can be transmitted to the specified location (registration) of the memory of another server application through external media ). If you want to use VI, applications, databases
Or the operating system must be from www.viarch.org


Obtain the corresponding API. Both DB2 6.1 and oracle8.1 use the VI architecture in their database cluster applications.
IP over Ethernet latency includes TCP
Stack (CPU load) and Ethernet transmission latency. The maximum transmission rate of 100 Baset is 100 Mbit/sec, And the FC-IP reduces Ethernet-related latency and delivers better throughput than IP over Ethernet at fiber-channel speeds, but still cannot avoid TCP
/IP software
Latency. The FC-VI removes TCP
Stack and provide DMA between application memory and HBA. The FC-VI has exceeded the system kernel, avoiding the context conversion and buffer changes of the operating system, achieving a higher transmission rate.
The FC-VI requires a fiber channel HbA that supports the VI architecture, the FC-VI HbA and the Fiber Channel HbA that supports scsi I/O are essentially different. Both Troika and finsar provide Fiber Channel hbas that support the VI architecture. Finsar sells a PCI-based fiber channel HbA that supports the VI architecture and supports point-to-point connection or switching. Troika sells a PCI-based smart controller-san 2000 series controller that supports FC-SCSI, FC-IP, point-to-point FC-VI, FC-AL and switching topology. Troika Controller provides multiple management options and features, such as protocol priority configuration and path transformation in Server Load balancer.
V. Multi-Protocol Structure
All of these three Protocols (FC-SCSI, FC-IP, FC-VI) can be combined into a fiber channel structure. While these protocols work in FC-AL, the corresponding bandwidth sharing and arbitration consumption ignores the performance advantages of FC-IP and FC-IP. Fabric exchange is recommended because fabric exchange provides multiple non-blocking 100 m/sec channels between servers and storage devices.

HDS developed a multi-protocol Business Intelligence solution that simulates the use of FC-SCSI, FC-IP, and FC-VI exchange structures
Solution. Using ancor, brocade, or using both fabric switches together, this solution was verified at the computing distributor exhibition in and the Gartner Group storage conference in. In this solution, fabric exchange integrates the centralized management of all data networks and provides unlimited Scalability (as servers and storage system nodes increase ). The use of fabric switching sacrifices some completely negligible performance loss, and in most cases, the consumption of FCP-SCSI via SWAp does not exceed 2%.

When more and more applications are being modified or developed for the VI architecture, we can see that more and more data networks are integrated into the fiber channel. Emerging market opportunities such as EAI (Enterprise Application exchange) need to provide real-time data transmission and exchange between different platforms and heterogeneous databases, FC-VI makes high-speed data exchange between servers possible. Existing applications such as NAS can use VI sockets to run their network files at the speed of Fiber Channel
System.
People are working hard to propose an IP standard for access to storage. Cisco has submitted a standard for SCSI over IP to IETF, which is still under development, it needs to separate control and command signals from data signal transmission cables, mainly considering the overhead of traffic control and transmission control.