Bracknell, Berkshire-based White Cross Systems Ltd says it will be shipping three of its new massively parallel, interactive database servers to Japan, and one to an unnamed London-based company before the year is out. The shipments are the first steps towards the company’s aim of bringing supercomputing power to the commercial workplace – to financial, […]
Bracknell, Berkshire-based White Cross Systems Ltd says it will be shipping three of its new massively parallel, interactive database servers to Japan, and one to an unnamed London-based company before the year is out. The shipments are the first steps towards the company’s aim of bringing supercomputing power to the commercial workplace – to financial, retailing, banking and insurance companies in particular – with its 9000 family of data servers announced last month (CI No 2,045). White Cross has gone back to the drawing board with its system, in recognition of the difficulties that established parallel processor manufacturers have faced trying to convert commercial applications to machines that were originally designed for technical and engineering applications.
The difficulty of exploiting true parallel processing lies in designing the software to resolve problems such as locking, optimisation and failure recovery without impeding performance, executive director Chris Delve told the audience at a recent product demonstration in London. After four years of research and development work, the company has come up with new hardware – machines that are fully optimised for relational database querying, without the need for an operating system and with no general-purpose file management system – together with closely integrated SQL-based database management software. It is marketing this combination as a black box – well dark blue box – that can then interface transparently with other familiar applications such as spreadsheet packages. Surrounding the core database software is a standard SQL 89 interface; Microsoft Corp’s Open Data Base Connectivity interface; and the International Standards Organisation’s Remote Database Access interface. The system is also directly connectable through TCP/IP protocols to Ethernet networks and can also operate on Open Systems Inetrconnection and Token Ring networks. White Cross says it will include support, where appropriate, for other standard Application Programming Interfaces and hopes to embrace the most popular local and wide area network channels in time. The current hardware is based on Inmos Ltd T425 Transputer RISC chips. These are rated at 10 MIPS each and are arranged in groups of six to give a total of 60 MIPS per board. Future White Cross products will be upgraded to the new 100 MIPS T9000 Transputer – without any major software re-work, White Cross says. Each chip is accompanied by 16Mb of memory, supplied by Hitachi Ltd’s 16M-bit zip Dynamic RAMs. This enables the system to hold most of the data in the memory rather than on disk, cutting the number of instructions and processor cycles involved and so speeding up data searches.
By Lynn Stratton
White Cross reckons that its entry-level 9010 can scan 1m 100Mb rows of data in 1.1 seconds – 0.8 seconds compile time, 0.3 seconds scan time. According to IBM Corp figures, a similar scan done on a mainframe running DB2, with sequential pre-fetch turned on, would take 83 seconds. This represents an improvement in price-performance of 130 times. White Cross’ systems have also come out well in a series of tests based on Set Query Benchmark, completing 50% of queries in under two seconds and a further 30% in under 10 seconds. This compares with highly variable response times for other database management software systems, most of which were over 10 seconds, with 41% over 100 seconds. This consistency is down to the multi-versioning data management method employed in the White Cross software. Here, updated rows do not over-write current rows, so that read-only processors always have easy access to a known and consistent view of the data. Disk storage is provided for fault tolerance using a RAID-3 algorithm. Here, if one disk fails, any data block can be constructed from the remaining three without the system grinding to a halt. In addition, a disk can be nominated as a hot standby, to replace another disk automatically in the event of failure.
In both the 9010 deskside server and 9020 data centre ser
ver systems, self-healing is provided by detecting the failure of sub-modules and reconfiguring to avoid the failed module. The deskside server should survive hardware and software failures with little disruption. The data centre server is fully fault-tolerant; it does not need to be powered down in order to replace a faulty disk and has a built-in uninterruptable power supply that can maintain service for ten minutes whilst a standby generator comes on line. Both systems are scalable. The 9010 entry-level system with minimum of two processor boards offers 120 MIPS and 192Mb RAM. This is expandable to a maximum of eight boards, delivering 480 MIPS and 768Mb within a single cabinet 30 high by 12 wide. The cabinet also contains 6.8Gb of disk space, 4.5Gb of which is available to the end user. The 9020 centre system can contain a maximum of five, three-processor racks with 19 processor boards each. This offers over 3,600 MIPS and nearly 6Gb of memory – a 700% improvement over the deskside server. Pricing for the data centre, which is still being tested and is due for release next spring, is unavailable. A 120 MIPS deskside system is UKP145,000. White Cross is hoping to sell around 50 systems worldwide next year and, to this end, is busy signing up distributors. The systems are assembled and tested in the UK.