TCPIP Storage Networking
TCP/IP Storage NetworkingThe advantages of TCP/IP storage networks is not only that you can use them to create a unified network but that the infrastructure and trained IT staff are more widely available and that Ethernet adapters, switches, and other hardware offer a great economy of scalethat is, components tend to either be less expensive to start out with or will become so over time. Storage networking over IP is leveraging the investment that the industry has made in Ethernet. File-oriented storage traffic already travels over IP networks; that's what NAS or filers are designed to do. But these new standards hope that all kinds of storage traffic, even block-oriented, DAS, or Fibre Channel SAN data will function over IP networks. Doing so will expose storage traffic to a whole host of IP-based applications (backup, replication, and so forth) while allowing IP management tools and staff trained in IP methods to manage storage assets more easily. Ethernet networks have always been used to send and receive storage data. Put aside the fact that NAS filers (NFS and CIFS) are often connected directly to a TCP/IP network, servers and clients are where very important data resides on DAS, and that means that backup traffic must flow from these systems. These historical uses of storage over Ethernet have been largely file based and have not been used to access block-oriented devices. The more recent developments, therefore, have been aimed at making storage traffic over TCP/IP a much more general phenomenon. It is expected that some of the technologies discussed in the next sections will have significant impact in the storage networking marketplace and will to some extent replace Fibre Channel SANs. There are three approaches to sending storage traffic over IP networks:
FCIPFibre Channel, while both scalable and reliable, imposes considerable cost. In addition, a significant amount of overhead is required to map Fibre Channel protocols so that they are compatible with their wrapper, and it is easy to transmit and translate the data when it arrives. When you exceed the distance that Fibre Channel can run (about 1.8km), storage traffic sent over long distances requires a different transport medium, and Fibre Channel is impractical. What you really want to do is leverage the availability of established IP networks to send Fibre Channel data. Therefore, the industry is developing new standards that allow for WAN transmission of Fibre Channel as well as allow Fibre Channel traffic to use IP LANs when Fibre Channel SANs do not exist. These FCIP technologies are also regarded as a means of solving the problem of connecting "SAN islands"that is, bridging two separate SANs by using FCIP gateways, whether they are located on different floors in a building, different buildings on a campus, or across the country. Fibre Channel and IP networking have two very different design principles. Fibre Channel was designed to be fast and highly reliable. IP networks were designed to be fault tolerant but not necessarily fast. That Quality of Service (QoS) difference greatly affects how you can use FCIP technologies. FCIP exposes this difference, especially IP's intrinsic latency. Many Fibre Channel SANs support high-speed transactional database systems. A transactional system like that can sustain a wait period of a second or so to retrieve some data in order to continue processing, but longer wait periods bring the transactional processing to a halt. When you send FCIP data across the room, there is perhaps one "hop": the latency involved with the path through the switch and the delay in its retransmission. Add a couple more hops to cross a city, and it's unlikely that FCIP is going to connect an OLTP database to primary storage. Some processes tolerate latency. Backup is one example of where FCIP could be used. In a backup process, data is transmitted from one site to another. If there is latency, the data throughput is simply slower. How much slower? At current speeds, data over an OC-3 IP connection takes about six times as long as over a Fibre Channel SAN. iFCP replaces the Fibre Channel transport layer with the TCP/IP protocol set and GigE wire transport. It does not create native IP packets, but it does allow for fast point-to-point dedicated connections between two iFCP gateways or switches. The iFCP gateways communicate with Fibre Channel devices and make the necessary conversion. The main market for iFCP deployment is enterprise-class backup, replication, and storage virtualization. iSCSIThe iSCSI protocol defines a set of rules that determine how to send and receive SCSI-3 commands along with the block storage data on SCSI attached storage over TCP/IP. Because what is being communicated is IP packets, you can send iSCSI over GigE. iSCSI is placed just above the data link layer in the OSI networking model, and it directly interacts with a host's SCSI access method command set. Although any TCP/IP transport is generally defined to work over any TCP/IP network and can even be implemented entirely in software, almost all implementations of iSCSI use a specially constructed HBA that interfaces with GigE. Not only can iSCSI connect two different IP SANs, it is also possible to use an iSCSI-FC gateway to provide access to storage on a Fibre Channel SAN. Thus iSCSI can link storage over not only LAN but MAN and WAN networks. However, the first iSCSI HBAs from companies such as Adaptec are aimed at getting SCSI storage that is DAS to be made available to other systems in small workgroup or departmental settings. We haven't yet seen enterprise iSCSI devices appear, but the HBAs that should enable this technology started appearing in the market in 2003. |
EAN: 2147483647
Pages: 240