PCIe adapters supporting long distance 10GB fiber?
denys at visp.net.lb
Tue Jun 20 21:07:39 CST 2017
On 2017-06-20 23:44, Baldur Norddahl wrote:
> But what foundation do you have for asserting that switch hardware is
> different in this regard? I can say that we are using 80 km modules in
> various hardware without any issues. I admittedly do not use any high
> modules in servers, but I will need better evidence than this to assume
> that it would not work just fine.
For switches i guess it is same story as for PoE on them - total power
So if you will pack whole EX4500 with 10G 80km SFP+ it might have
problems as well,
but for normal use, and if few only are "long distance/high power", at
any case 3.3V supply rail
by design in switch should handle many SFP, so if there is 48 ports, it
should handle by specs at least 72W peak load.
It might be multiple power rails for groups of ports, but still, much
better than just 750mA on network card.
But that's just guessing, i never seen circuit diagrams of good
switches, or at least reference design,
as it is all NDA material.
> Den 20. jun. 2017 22.24 skrev "Denys Fedoryshchenko"
> <denys at visp.net.lb>:
> On 2017-06-20 22:07, Baldur Norddahl wrote:
>> I would expect anything mounted in a computer to have all the power
>> could want. It is not like the ATX power supply cares about an extra
>> or two.
>> As I understand the issue it is more about cooling than power and is
>> primarly a concern in high density switches were you could have 48 or
>> to power and cool.
> SFP needs 3.3V, it might be supplied from regulator on the card or
> can't be absolutely sure, in reference design it is just 3.3V_NIA and
> also reference design SFP power circuit define max 750mA/3.3V max to
> thats only 2.475W.
> FTLX1471D3BCV (10km SM) - up to 285mA
> FTLX1671D3BCL (40km SM) - up to 400mA, and thumb rule in electronics it
> better to not exceed 50%
> of max specs of designed max current, as for many parts it is stated
> 25C & etc operating conditions.
> I expect it might work, but noone knows how long, and how reliable, if
> is not cooled very well.
> And 82599 sensitive to cooling(it is very old card after all), as soon
> it is not enough, it starts to glitch.
>> Den 20. jun. 2017 18.09 skrev "Denys Fedoryshchenko"
>> <denys at visp.net.lb>:
>> I guess it depends on NIC, there is many spinoffs of Intel X520 with
>>> weaker power supply circuitry.
>>> It might work with good NIC, but you can't rely on it on long term,
>>> Even 40km Finisar SFP+ has Pdiss 1.5W. Also they mention: "The
>>> power consumption of the FTLX1672D3BTL may exceed the limit of 1.5W
>>> specified for the Power Level II transceivers"
>>> If we talk about 80km, Pdiss is 1.8W.
>>> While 10GBASE-LR is <1W
>>> On 2017-06-20 16:30, Max Tulyev wrote:
>>> We use Intel NICs with SFP+ holes. It works good with long and short
>>>> range SFP+ modules, including CWDM/DWDM.
>>>> On 15.06.17 12:10, chiel wrote:
>>>>> We are deploying more and more server based routers (based on BSD).
>>>>> have now come to the point where we need to have 10GB uplinks one
>>>>> devices and I prefer to plug in a long range 10GB fiber straight
>>>>> the server without it going first into a router/switch from vendor
>>>>> x. It
>>>>> seems to me that all the 10GB PCIe cards only support either copper
>>>>> 10GBASE-T, short range 10GBASE-SR or the 10 Km 10GBASE-LR (but only
>>>>> few). Are there any PCIe cards that support 10GBASE-ER and
>>>>> 10GBASE-ZR? I
>>>>> can't seem to find any.
More information about the NANOG