PC Servers Grow Up With Superspeed
Until very recently, PC servers lacked the dedicated I/O subsystems, dedicated I/O processors and many of the other dedicated accoutrements of their high-end Unix, minicomputer and mainframe counterparts. In PC operating systems such as Windows NT 4.0 and Windows 2000 – both of which rely extensively upon virtual memory and upon the ability to move data into and out of memory very quickly – this situation traditionally posed a significant shortcoming.
Enter SuperSpeed, a software utility from Superspeed Software Inc. (Sudbury, Mass.) that purports to eliminate the PC server's most notorious bottleneck: Its high disk access latency and poor – comparatively speaking – disk I/O throughput.
Server performance is all about throughput and bandwidth. The Pentium III microprocessors that anchor most of today's PC servers are designed around a bus topology that can deliver a maximum memory bandwidth of 1.3 GBps—a pretty fat pipe for single- to quad-processor SMP boxes.
In stark contrast, and even when fed by a fat fibre channel pipe, the best that today's disk I/O technologies can do performance-wise is somewhere in the neighborhood of 125 MBps—and that in sequential or same-sector I/O transfers. Random I/O transfers – which form the bulk of most disk-related activity – are only about 1/5 as fast.
SuperSpeed-NT and SuperSpeed 2000 – for Windows NT 4.0 and for Windows 2000, respectively – effectively move an entire disk partition or an entire database into system RAM. The catch with either product is, of course, that the partition or database in question must be small enough to fit into available system memory in the first place.
In Windows NT 4.0, SuperSpeed-NT is limited by the maximum amount of physical memory that the operating system itself can address – 4 GB. In Windows 2000, which features support for the 36-bit Physical Address Extensions that Intel Corp. (Santa Clara, Calif.) introduced with the Pentium III Xeon, SuperSpeed 2000 has considerably more memory to play with – to the tune of 64 GB.
Superspeed claims that its disk caching technology can improve overall system performance by almost 3000 percent. Because the application or database that resides in memory operates at CPU and memory speed, it remains unencumbered by the trade-offs associated with disk caching and with disk I/O. For I/O intensive applications – which can often suck precious cycles away from a host CPU – Superspeed representatives claim that similar performance benefits can accrue.
"SuperSpeed 2000 and SuperCache 2000 ... [enable] users to increase the performance and scalability of Windows 2000 by eliminating slow disk read/write operations," he explains. "Windows 2000 customers are now able to extract peak performance from all their applications."
SuperSpeed uses available physical memory as a virtual disk or as a kind of RAM disk. When a system reads an area of data or code, the data is returned at fast memory bus speed - theoretically eliminating the latencies associated with disk mechanical actions, host bus adapter overhead, I/O bus overhead and the low level drivers associated with disk access. Company representatives claim that Superspeed can increase effective I/O performance for both simultaneous users and for batch jobs on a system by a factor of over 100 -- depending on the number of CPUs in a system and on the amount of free memory available to SuperSpeed.
SuperSpeed will run on vanilla Windows NT 4.0 or Windows 2000 systems. Depending on system memory size, a SuperSpeed RamDisk can be up to 28GB in size.
When all is said and done, the folks at Superspeed could be on the leading edge of an emerging paradigm in the mid-range computing space.
According to Mike Casey, vice president of marketing with solid state storage specialist Solid Data Systems (Santa Clara, Calif.), more and more IT organizations are turning to solid-state storage solutions in an attempt to address nagging performance issues. Casey says that the low-latencies and incredible speeds that such solutions typically provide can be particularly attractive in transaction-oriented e-commerce environments.
"If it's transaction oriented, [solid state storage] is great. If you're going after small random reads, or temporary workspaces for joins and queries, that's where putting that kind of work on solid state disc gets your server out of the wait-for-I/O mode," he explains.
While not a solid-state storage solution in the classic sense of the term, Superspeed does rely upon essentially the same model: Frequently accessed or transaction-oriented data is moved into system memory, from which it can be more efficiently accessed and updated.
Microsoft itself has shown an interest in this type of technology. One of the last features yanked from the Windows 2000 Release Candidate was a so-called "In-Memory" Database (IMDB) that provided an additional layer of cache memory which could theoretically enhance performance between the operating system and stored data to increase performance for OLE DB-compliant data stores.
Like Superspeed, Microsoft's short-lived IMDB achieved this effect by caching information at the middle tier - in physical memory -- and by eliminating the need to go back to disk to access certain data.