How do I determine the I/O bandwidth capacity of a configuration like the one Bhaskar described in his demo for 1 million IOPS? Using 4K blocks, it looks like he achieved 4GB/sec or about 50% PCe 2.0 1x16 would give 16*500MB/sec or 8GB theoretical. Ignoring the OS impact, am I correct in believing that his config is limited to 8GB/sec?
His test is described at http://communities.intel.com/community/openportit/server/blog/2009/09/26/the-one-million-iops-game
There's a YouTube video too. Sorry if I have the wrong group - bandwidth is a complex issue including many factors. Please advise if I should direct the question elsewhere.
Impressive! I don't know if you'll get many responses here, but I see you should be able to comment on his blog directly. Looks like he's taken time to reply to those interested. His last post was only five days ago.
Apologies for late reply, I don't frequent these forums very often and didn't see your post.
My solution was driving 4Gigs/s worth of data using all 5 PCIe x8 slots, its not a consumption of lot of BW by PCie standards, but yes theoretically you can drive 8GB/s (4GB/s one way) data on a x16 PCIe. I could possibly test this with larger chunk size read/writes and see maximum amount of data transfer i can achieve within SAS/SATA limits (3Gb/s per port). A modern graphics card might be a good candidate to test potential of a PCIe bus.