- May 29, 2011
Well the big advantage to fiber is that it actually works, if you play by the rules, without drama even. 10G Category cable is subject to lots of drama because it burns more watts, is higher latency, and has tragic length limits.Thank you for the responses and information.
That seems doable. I've been Googling and I'm assuming I need OM4 Fiber Optic cable then some SFP+ connectors/wall jacks? That was the downside in my mind of running SFP straight to my desktop, as it's about a 30 foot run and would require getting another new card for the desktop.
Is there a particular advantage to doing things this way versus swapping for an RJ45 Intel (for the NAS), then picking up that 10gbe switch?
Of course I'm impatient and already ordered 50 feet of cat6a, connectors and installation tools last night :p
10G SFP+ is a little nasty for beginners because it seems so "foreign", what with no RJ45's, and these weird SFP module thingys, and a bewildering array of connectors and types of fiber, which is what caused me to write the 10G Networking Primer. But the cold truth is that this stuff has been successfully deployed in data centers for the last SEVENTEEN years. By way of comparison, I wrote back in 2014 that I felt that 10G copper was about to make a big entrance, and would ultimately prevail because it was "simpler", despite being a lower quality (think: "VHS of the networking world") product. Here I am five years later, pleasantly wrong-ish. There's still very few cheap 10G copper switches, little overall adoption, etc.
Perhaps this makes sense. Copper's a crappy technology.
10G copper chipsets typically require at least 3W to sustain a connection, and this can go as high as about 10W to drive a long run. By comparison, SFP+ ports are spec'd for 2.5W *MAX*, normally using less than 1W, and with that, I can put in an SR (think: "short range") optic and with OM4 I can run as far as 400 meters without a problem. Or I can put in an LR (think: "long reach") optic for a run up to ten KILOMETERS.
10G copper chipsets normally introduce around 3 microseconds of latency in processing, whereas SFP+ is typically around 300 nanoseconds, or about 1/10th as much, making SFP+ a better choice for performance.
Is somewhat dated at this point but still generally correct-ish.
But I think the real suppressor is that there just isn't much demand. When we reached 1Gbps, this was really sufficient for most general end-user computing needs, and because it took a decade for 10G copper to evolve, people had a decade in which to appreciate that they weren't really being killed by the constraint of 1Gbps, and for dirt-cheap 1G switches and cards to be produced. I can create a cheap 1G network with two 1G ethernet cards and a switch for about $50. Now try that trick with 10G. Most people simply won't swallow the cost differential. Even here, the core network is all 10/40G, but end user machines are 1G.
Anyways, probably more than you wanted to know. What you DO need to know:
You don't need OM4. OM3 is good enough. But OM4 is dirt-frickin-cheap too.
As an example, FS.COM 5 meter OM3 for $4.30. You can order a wide variety of prebuilts, which sometimes aren't actually in stock, so be looking at the ship date. You can choose OM3, OM4, the length of your choice, make sure to select LC for both ends, select an appropriate jacket (depends if you're running in-wall or have fire code requirements), etc. I suggest the thick 3mm stuff for beginners. I do this stuff professionally and the 0.9mm stuff still freaks me out a bit.
That last one is OM4 BIF duplex "uniboot" with a 2mm jacket. Crazy stuff. We're switching over to it, all ordered custom length. Very cool, and really not that pricey.