Here’s some RAW SAN network speeds that I found in some post somewhere (which I didn’t write down).. Obviously there are a lot of caveats related to this, but from a pure bandwidth perspective, I thought this was interesting for reference.
1 gig = 125 MB/sec
2 gig = 250 MB/sec
4 gig = 500 MB/sec
8 gig = 1000 MB/sec
10 gig = 1250 MB/sec
This is just a basic bits-to-bytes conversion (8-bits/byte) and you’ll never see speed at those numbers as there will always be protocol overhead, etc.
As to where it came from, it sounds like the usual vendor marketing fluff. ๐
Thanks for the comment Andrew. And you’re totally right, thus why I made the “caveat” comment in the post. Protocol overhead/type of protocol, array caching architecture, Raid overhead/type, all of these affect these numbers. These numbers are purely from a wire speed point of view. It seems some of the storage vendors we deal with as of late have been claiming “wire speed” performance on their arrays. Well, #1, we can see what wire speeds are based on the simple math on this post. #2- We know that protocol overhead will subtract from your performance, which #3- if the vendor’s claims on performance are too close to the numbers above, then they are doing either some awesome caching or are are flat out fudging their numbers..
I’ve always believed the 10b/8b encoding that occurs when transmitting bytes serially neutralized this argument. Unlike the good old days of parrallel cables ๐ when the parity was sent in parrallel with the data bits, we now have to incorportate them into the serial data stream, e.g., GigE or FC, and use 10b/8b to achieve a similar result with regard to parity. Said another way, you need to send 10 raw bits to end up with 8 useful bits on the other end (SERDES) http://en.wikipedia.org/wiki/SerDes. Doing the math then, take the 1Gb = 125MB/s and back it off to 8/10’s of the number and your back to 1Gb = 100MB/s.