Silicon Investor (SI) -- The First Internet Community

STOCKTALK

We've detected that you're using an ad content blocking browser plug-in or feature. Ads provide a critical source of revenue to the continued operation of Silicon Investor. We ask that you disable ad blocking while on Silicon Investor in the best interests of our community. If you are not using an ad blocker but are still receiving this message, make sure your browser's tracking protection is set to the 'standard' level.

Technology Stocks : Compaq -- Ignore unavailable to you. Want to Upgrade?

To: John Koligman who wrote (70426)	10/27/1999 11:17:00 PM
From: rudedog	Respond to of 97611

a 32 way wildfire could directly connect 512 "devices" - like fibre channel controllers, disk array controllers, etc. In the ENSA architecture, things like storage are fibre-attached, and each attach point can have 1024 storage subsystems, although much smaller numbers - 8 or 10 - are more common. The storage subsystems consist of intelligent disk array controllers and fairly large numbers of physical devices. Typical configurations would be redundant fibre links to multiple hosts and multiple storage subsystems. The approach is different than S390 concepts but similar in that the grunt work of managing the individual devices is pushed way down - so an array of 15 disks might appear as a single big drive (hundreds of GB maybe) to the host systems, but with very high bandwidth and very low access times. Access time for large data blocks is divided by the number of physical devices in a stripe set. Access times in the hundreds of microseconds can be achieved, and large cache in the subsystem can reduce that even more, to the tens of microseconds. Even a gigabit fibre link is easily saturated at these data rates, so the multiple links work in parallel in the absence of a link failure to provide greater bandwidth. Failure modalities are managed in a way that would be very familiar to an old mainframe hand - highly available configurations use process pairing to achieve sub-second failover to recover transparently from failures of processor or memory subsystems, multiple paths assure reliable data delivery, and the storage and I/O subsystems, including network links, also recover transparently and without any special programming - those recoveries happen below the OS and are invisible to the host. If the customer wants to pay for it, multiple levels of failure can be recovered automatically by replacing failed resources with available free resources - network, storage, even processor and memory, to maintain fault tolerance even after a failure. I used to do a lot of work on high-reliability S390 configurations (it was a ways back) and the concepts don't change much, but the architectural support is a lot different. I think the new stuff is pretty slick...