6 Conclusions

 

In this paper we propose a new device : the Network RamDisk. This is a block storage device that consists of (idle) main memory of workstations within a (heterogeneous) workstation cluster. We describe our prototype implementations of a Network RamDisk implemented on top of the Digital Unix 4.0, and the Linux operating systems as a device driver. No modifications were made to the kernel of the Digital Unix 4.0, or the Linux operating system.

The contributions of this paper are:

• We describe how to build a remote memory disk storage system. The major advantage of our design is the portability it offers. Our device driver can be installed on any Digital Unix or Linux system, and can exploit the main memory of any workstation supporting TCP/IP.
• We propose a new adaptive parity reliability strategy for distributing disk blocks to many servers. Studying the performance of our policy, we show that it effectively reduces the time for redundancy cleaning, thus providing reliability at a very low performance cost.
• We show that storing disk blocks to remote memory results in substantial performance improvements over the local magnetic disk, mostly due to latency reduction.
• We provide a new efficient way of making remote memory available to every application through the common filesystem structures.

Based on our implementation and our performance results we conclude:

• Storing data to the Network RamDisk results in significant performance improvement over storing them on the traditional magnetic disk. Applications that use our Network RamDisk even when running on top of traditional Ethernet technology show performance improvements of up to 412% (Figure 8 ).
• Organizing remote memory as a Network RamDisk under any ordinary filesystem, is an inexpensive way to let applications use the memory of any workstations in a heterogeneous cluster. Everyday applications can now use remote memory for their data files without modifying their code, under any Unix filesystem.
• Remote memory data storage provides good performance and reliability with almost no extra hardware support.
• The benefits of storing data blocks to remote memory will only increase with time. Current architecture trends suggest that the gap between processor and disk speed continues to widen. Disks are not expected to provide the latency needed by several applications unless a breakthrough in disk technology occurs. On the other hand, interconnection network bandwidth keeps increasing (and network latency decreasing) at a much higher rate than (single) disk bandwidth, thereby increasing the performance benefits of the remote memory disk storage.

Based on our performance measurements we believe that using remote memory as a Network RamDisk is a cost-effective and performance-effective way to execute I/O intensive applications on a network of heterogeneous workstations.