Sequent Computer Systems Inc says Sun Microsystems Inc’s view that in a ccNUMA system all memory should be created equally simply does not represent the facts (CI No 3,601). Memory is not created equally, it says. Memory which holds instructions that are needed often should be treated differently to memory which holds instructions that are only used once. Why would anyone want to be able to access data and instructions that are needed only very occasionally as quickly as instructions that are needed all of the time? ccNUMA acknowledges this relationship. Sequent says that as memory latencies – the time penalty for having to fetch instructions from remote memory rather than local memory – are halving each year, the whole latency issue is going to fade away quickly. Sequent says that its takes three or four times as long to reach remote memory as local memory in its NUMA-Q ccNUMA servers. Between 90% and 98% of all memory access is to local memory. Sequent says the only ISVs it needs to work with to have them support its architecture directly are heavy users of middleware, such as Oracle. Baan, Peoplesoft and other applications run unchanged, it claims. Remember, it says, it introduced ccNUMA as the logical way to link Intel’s SHV four-way boards. Sequent is currently building a campaign around the ability to run Unix and Windows NT on the same Numacenter servers but admits it’s currently more of a framework notion than a product play. When Intel and Microsoft ship 64-bit products, that’s when the notion gets interesting, Sequent says. It believes ccNUMA is currently way down on Microsoft Corp’s list of concerns. Sequent says it’s clear Microsoft is encouraging lots of partners to experiment with NT and ccNUMA – Sequent, Silicon Graphics and Data General to name a few – and show it the fruits of their labors. However until Microsoft either endorses one approach or another and implements ccNUMA much of the discussion is a moot issue. Sequent expects that Microsoft will integrate support for a ccNUMA model and leave the hardware vendors to implement it. In June, Sequent is to enhance its system configuration options, as part of which it will enable full memory partitioning.
