Thanks for the Banderacom FAQs post. It caused me to re-read a December article about WIND’s partnership with Banderacom. In particular, the following highlights are interesting and deserve comment: (bold emphasis by me)
InfiniBand communicates like a packet-based network. Rather than relying on a processor, as does PCI, InfiniBand employs intelligent I/O engines,, which are referred to as channels. A series of interconnected channels form a “fabric,” which physically combines connectors, cables and switches using one or more host channel adapters.
Initially, InfiniBand’s low-hanging fruit will be for server clustering, Grove said, and for high-speed server I/O, and server-to-server and server-to-storage communications. “There’s an opportunity to build all kinds of other InfiniBand equipment to connect all these clusters of servers together and a variety of target equipment for storage, LAN and WAN interfaces.” The current specification supports 12 channels, which may be combined to provide as much as 30GB per second of bandwidth, Grove said.
Banderacom initially sought to partner with Wind River (www.windriver.com) due to popular demand. “We did a survey of our customers and found that a majority of them were going to be using a Wind River RTOS on their target applications,” he said, adding that InfiniBand also was a market that Wind River was interested in getting into. The company is scheduled to begin its first shipments of the iBandit InfiniBand SoC in mid-2001.
The packet-based network design of IB that requires “intelligent I/O engines” is what makes WIND’s ownership of the I2O iRTOS space so important in getting IB off the ground. Technically I2O is not a mandatory requirement of IB, but from a practical point of view it, or something very much like it, is. This clearly is a major reason why WIND showed up so dominant on the survey that the partnership between the companies must have seemed preordained.
WIND investors have struggled for years (since the start of this thread to be precise) with the unfulfilled promise of I2O and, over the last couple of years, NGIO which evolved into IB. Fortunately, there are plenty of other lily ponds to feed WIND outsized profits while IB matures. A prize this big is not available for easy plucking. Otherwise, a company with deep pockets would simply push WIND aside and dominate the IB intelligent I/O landscape. The survey and now the partnership with Banderacom point to already intense network effects, in the form of proven capability, interoperability and familiarity, developing around WIND’s iRTOS solution – an outcome of years of intense joint development by WIND, Intel and other major partners. Hard-earned network effects forms the major barrier that effectively protects WIND’s impressive dominance in the IB iRTOS space.
The server appliance lily pond exists today mainly because of I2O, and with or without IB, could skyrocket in 2001. The catalysts might well be I2O-based, off-loaded LAN NICs, called iLAN and most appropriate for Gigabit LANs, and a Linux-based network storage device that implements iRAID and iLAN. Wouldn’t it ironic if the server appliance lily pond explodes before IB even gets a chance to contribute its enormous potential?
While I can’t pinpoint the timing for the full realization of these developments, they do seem to be, once again, just around the corner.
Allen |
