http://www.denali.com/wordpress/index.php/dmr/?disp=comments en-US http://backend.userland.com/rss 60 Tue, 09 Feb 2010 21:54:31 +0000 rd4sndk [Visitor] c528@http://www.denali.com/wordpress/ Two probables with your analyst. <br /> <br /> 1. PCIe throughput is higher because its a parallel interface that works for DRAM or SRAM (onboard memory) but not for mass NVM storage. One will need thousands of drives and as such would require multiple PCIe cards most of which will have to be located off the server. transfering data from a standalone storage device will require a serial interface and that will be one of the standards your article is negative towards. <br /> <br /> 2. I very much doubt after the first enterprise PCIe failure that an operation will continue to place NVM storage without multiple redundancy paths and hot swappable capability as there main NVM storage solution. So shutting down the server to replace a PCIe card wont fly.<br /> <br /> Long term, PCIe may represent an accelleration storage device sold in pairs. Above that would requirement nothing else makes a lot of since. Given that assumption, PCIe will represent at most 10% of storage going foreward. I think SAS will dominate because of its bidirectional capability. As a novice, I fail to see the need for SATA in enterprise storage but that's because I don't know enough about either SAS and SATA. <br /> <br /> Good luck with that PCIe interface investment. I sure would like to see a petabyte system built around PCIe. The downtime of such a system would clobbler an enterprise operation. Even if you can make a system were its hot swappable you still have the issue of driving a parallel bus off board. I personally would go down that path.
1. PCIe throughput is higher because its a parallel interface that works for DRAM or SRAM (onboard memory) but not for mass NVM storage. One will need thousands of drives and as such would require multiple PCIe cards most of which will have to be located off the server. transfering data from a standalone storage device will require a serial interface and that will be one of the standards your article is negative towards.

2. I very much doubt after the first enterprise PCIe failure that an operation will continue to place NVM storage without multiple redundancy paths and hot swappable capability as there main NVM storage solution. So shutting down the server to replace a PCIe card wont fly.

Long term, PCIe may represent an accelleration storage device sold in pairs. Above that would requirement nothing else makes a lot of since. Given that assumption, PCIe will represent at most 10% of storage going foreward. I think SAS will dominate because of its bidirectional capability. As a novice, I fail to see the need for SATA in enterprise storage but that's because I don't know enough about either SAS and SATA.

Good luck with that PCIe interface investment. I sure would like to see a petabyte system built around PCIe. The downtime of such a system would clobbler an enterprise operation. Even if you can make a system were its hot swappable you still have the issue of driving a parallel bus off board. I personally would go down that path.]]> http://www.denali.com/wordpress/index.php/dmr/2010/02/09/the-evolving-enterprise-ssd-gartner-s-fo#c528