4.4 Building Blocks of Distributed Systems

Section 4.3 outlined several advantages for using network connectivity to deliver storage services. As shown, the only true means to deliver scalability is through the use of distributed systems across some type of networked infrastructure. Recognizing this fact, we examine how some of the fundamental thinking behind storage has changed.

The basic building blocks of distributed systems go back to early computer design and the separation of primary computing functions such as memory, CPUs, storage (I/O), and networking. Each function serves a distinct purpose yet is an integral part of the complete system. The basic functions are shown in Figure 4-12.

Figure 4-12. Primary computing functions.

For quite some time, memory and CPU capabilities and speed far outstripped that of I/O and networking. The doubling of processor transistors every 18 months, as identified by Moore's law, and the rapid advances in memory density kept these components ahead of the relative technology curve. The most frequent bottlenecks to system processing stemmed from the I/O and networking components . This led to server-centric solutions in which I/O and networking needs were sacrificed in order to enhance overall system performance. For example, the solution may have relied only upon DAS to fulfill throughput needs. Alternatively, network replication may have been restricted to overnight hours due to limited bandwidth. This gap is shown in Figure 4-13.

Figure 4-13. The gap between computing resources.

Following the silicon chip lead established by processors, complementary advances in networking technologies helped level the playing field across computing resources. Within the last five to ten years , interconnect technologies such as Fibre Channel and Ethernet have reached speed and performance levels that more closely match those of CPU and memory capabilities. In addition, the lines between I/O and networking are getting blurrier through the proliferation of NAS and block IP storage technologies.

These factors contribute to two shifts in thinking about networking and storage strategies. First, I/O and networking have outgrown previous performance limitations. Second, the underlying technologies are converging to the point where they can now be treated more equally. This shift is outlined in Figure 4-14.

Figure 4-14. I/O and networking increase speed and converge architecturally.

Equal treatment of storage and networking brings us back to the original breakdown of core computing functions ”memory, CPU, I/O, and networking. The most scalable and efficient computing solutions are those that effectively balance and utilize these resources. Distributed systems enable these kind of solutions. The key to going forward is to determine the optimal methods of building the larger distributed system, the communication between the resources, and a means for overall management.

Since networking is the essential glue for any distributed system, we take a closer look at how network control points impact storage networking and I/O in Section 4.5.