QSFP: Difference between revisions

From Helpful
Jump to navigation Jump to search
mNo edit summary
mNo edit summary
 
(2 intermediate revisions by the same user not shown)
Line 1: Line 1:
{{#addbodyclass:tag_tech}}
{{stub}}
<!--
<!--


From the perspective of switch/NIC hardware, QSFP is a physical slot that leaves the PHY translation up to the thing that slots in (as opposed to switches that have the PHY type bolted in - this may be a little more cost-effective, but less flexible).
In the world of network switching hardware (and to a lesser degree host [[NIC]]s), '''SFP''', '''QSFP''' and acronyms like it
 
describe a kind of physical slot that leaves the lowerst level of the [[network stack]],
QSFP is also hot-swappable (at physical layer, not all ''uses'' will be equally happy with you doing that).
the PHY translation, up to the thing that slots in.
 


The switch-transciever connection is OSI layer 1 - they move data, they are protocol-agnostic (they don't inspect it).
The switch-transciever connection is OSI layer 1 - they move data, they are protocol-agnostic (they don't inspect it, they just move it).


This means that with the right transceiver, it supports
This means that with the right transceiver, it supports
Line 17: Line 19:




10GBASE-X: offers choice between long distance and cheap at short run,
'''A little more practically'''
CX4: Copper (4 parallel paths), ~15m over infiniband-style cable -- and a lot cheaper than LX4
LX4: Optical (4 wavelengths of light), depending on optics ~300m or 10km
...but will probably be replaced by 10GBASE-LRM for long distance optical, and 10BASE-T (Cat & RJ) for short runs.


In hosts there is not a lot of difference between
: a SFP card plus specific PHY, and
: a card that happens to have that PHY on its board,
but in network switching hardware, the added flexibility (at some extra cost) can save so much headache and, in ''some'' situations, money.
Around switches, seeing the acronym may mean "aside from 1gbit or 10gbit ports, it also has one or two SFP style slots that leave it open how you connect that to the rest of your infrastructure" (e.g. optical/copper variants of 10GBASE or faster, 40gbit)
Only in a limited number of cases would you want to opt for a switch that consists of ''primarily'' unpopulated SFP style slots.
QSFP is also hot-swappable (at physical layer, not all ''uses'' will be equally happy with you doing that).
On the hardware side


* SFP (Small Form-factor Pluggable)
* SFP (Small Form-factor Pluggable)
Line 41: Line 54:


* QSFP+
* QSFP+
** Current max 112Gbit/s (4x28Gbps)
** Typically 40Gbit max, mainly aimed at four  10 Gbit/s channels, 10GFC FiberChannel, or QDR InfiniBand, sometimes a single 40GBit Ethernet link
 
 


50 Gbit/s QSFP14
* QSFP14
:: 50 Gbit/s, aimed at FDR InfiniBand, SAS-3, 16G Fibre Channel


100 Gbit/s QSFP28
* QSFP28
:: 100 Gbit/s, aimed at 100 Gigabit Ethernet, EDR InfiniBand, 32G Fibre Channel


200 Gbit/s QSFP56
* QSFP56
:: 200 Gbit/s, aimed at 200 Gigabit Ethernet, HDR InfiniBand, or 64G Fibre Channel





Latest revision as of 18:47, 22 April 2024

This article/section is a stub — some half-sorted notes, not necessarily checked, not necessarily correct. Feel free to ignore, or tell me about it.