Why would I choose Copper over SFP+ for 10GbE?

Solution 1:

The latency is basically negligible. The 10GBASE-T has a latency of <1microsecond. SFP+ has less latency in itself, but SPF doesn't include some of the physical transceiver (which may or may not add latency); hence the need for a physical module (or direct copper cables).

The biggest differences are price, as you've noted, and distance. SFP+ Direct Copper the cables have to be <15m (10m for certain cables). 10GBASE-T goes the standard 100m. Cat6 cabling is quite cheap (compared to other 10G cabling), and I would suspect that the equipment manufacturers "make up for that" in the price in addition to 10GBASE-T not being as popular yet.

The 10GBASE-T standard also uses more electricity (which causes both the increased distance and latency). The extra amount used isn't normally a factor.

Solution 2:

"SFP+ has a lower latency" due the much better noise/interference isolation, but copper is generally less fragile/more durable (if it will be in a "higher chance of being re-handled" environment).


Solution 3:

The 802.3an (10GBase-T) standard calls for latency of 2.5 microseconds or better. You are dealing with storage that still has latency measured in milliseconds. The difference might matter for extremely specialized high-performance computing applications, it can't possibly have any significant impact on your SAN performance.

How does the cost difference look after you have purchased the actual cables? You may find that market prices for CAT6A patch cables are becoming almost similar to prices for CAT5e, whereas the SFP+ cabling could actually be a significant component of your project cost. (It may even cancel out the difference in switch prices.)

I would suggest that overall project cost is likely to be the most important deciding factor.

(Disclaimer: there is no 10GbE at all in my current environment.)


Solution 4:

Here is a couple of links talking about latency with 10GBase-t. Basically 10GBASE-T is slower than 1000Base-t or gigiabit for small packets. If you are doing something like iSCSI then it will be insignificant, but if you are doing hundreds of thousands of short key/value lookups between servers that traverse several switches it can be significant and surprising that it's slower than gigabit...

http://www.datacenterknowledge.com/archives/2012/11/27/data-center-infrastructure-benefits-of-deploying-sfp-fiber-vs-10gbase-t/

http://www.plxtech.com/files/pdf/support/10gbaset/whitepapers/10GBase-T_1000Base-T_Switches.pdf

Note: The latency is from 10GBase-T, not copper. If you do SFP with integrated twinax cable, that is copper but doesn't have the latency problems of 10GBase-T.


Solution 5:

Comparing SFP+ to 10GBASE-T.

Pros of SFP+

  • Lower latency (though unless you are in HFT or similar it's likely negligable)
  • Lower power consumption
  • Cheaper NICs and switches
  • More choice of connected equipment.
  • With transcievers and fiber basically any run length can be covered.

Cons of SFP+

  • For short simplw runs needs "direct attach" cables which cost substantially more than twisted pair and are considerbally more cumbersome.
  • For longer runs or runs that need to go through patch panels needs transcivers and fiber. Fiber itself is cheap but transcievers, termination, patch panels etc for fiber an get quite spendy.

Pros of 10GBASE-T

  • Cheap twisted pair cables
  • 100m runs without messing arround with transcievers
  • Patch panels can be used without messing arround with transcievers.

Cons of 10GBASE-T

  • Higher power consumption
  • People may get tempted to use substandard cabling, just because it's twisted pair doesn't mean you can be anywhere near as sloppy as at lower speeds.
  • No good way to extend length beyond 100m (though this can be somewhat mitigated by chosing switches with mostly 10GBASe-T but also a handful of SFP+ ports)
  • limited choice of equipment.

Until/unless they can get the price and power for 10GBASE-T down it IMO has fairly limited utility.