Quote:
|
Originally Posted by duke900ssd SVC is a small system solution.
It only supports a small number of connections (I/O bottleneck) and introduces a PC (OMG!) between your enterprise servers and SAN.
Who would / could possibly consider such a solution in an enterprise class infrastructure? |
It sounds like you don't know too much about SVC. Small number of connections? No, not at all. A full SVC will support non-cached 960,000 IOs per second. Hardly a small number of connections, huh?
Each node is an IBM xSeries server - not nearly he same as a PC - running a specialized Linux kernel. Node are deployed in highly available node pairs called an IO group. In the even of the failure of one node, the other node in the IO group takes over all I/Os. If you loose a IO group, the IO is equally divided among all IO groups.
Each node has its own dedicated (and required) UPS.
Keep in mind that the SVC functions as an in-band controller. You don't cable anything directly to the SVC. Traffic is routed through the SVC via SAN zoning and makes your switch zoning much simpler to manage.
SVC introduces no IO latency. To the contrary, the additional layer of caching that SVC introduces typically improves IO.
Please have a better idea what your speaking about before giving bad information. There is an automaker in Michigan - can't mention which but there aren't too many to choose from, running an SVC with a IBM provided code modification. Their SVC is 40 nodes and they get 4.4 million IOs per second with 7 petabytes of managed heterogenius storage.
You were implying something about performance, scalability and reliability?