On Wed, Mar 28, 2018 at 9:44 PM, Paul Mather <p...@gromit.dlib.vt.edu>
wrote:

> On Mar 27, 2018, at 8:10 PM, Moshe Katz <mo...@ymkatz.net> wrote:
>
> Many thanks for the information and advice.  It is much appreciated.
>
> > According to the specs that I found on HP's website, your HP switch does
> > not support 10Gb, only 1Gb on its mini-GBIC ports. You will definitely
> need
> > a new switch to take advantage of 10Gb.
>
>
> It's true that the mini-GBIC ports support only 1Gb, but that HP switch
> also can accommodate two(?) option modules at the rear of the switch that
> can be used to provide 10 Gb connectivity.  According to the "HP ProCurve
> Switch - What modules are available for the 2910al?" page at the HP Support
> site (https://support.hpe.com/hpsc/doc/public/display?docId=emr_
> na-c02620659 <https://support.hpe.com/hpsc/doc/public/display?docId=emr_
> na-c02620659>) you can get dual-port 10-GbE CX4 and 10-GbE SFP+ option
> modules.  Well, at least I suppose you could, as I'm not sure how widely
> available they are, and, this being an old switch, it may be that buying
> option modules from vendors with which $WORK are willing to purchase may
> result in them being prohibitively expensive due to them being
> legacy/discontinued equipment.  In my experience, those tend to command a
> premium price (except when buying via eBay).  (The SFP+ option
> module---J9008A---appears to cost $700+ on a quick search.)
>

I missed that in the documentation, but something tells me you probably
don't want to do that anyway - with hardware that old that uses a
proprietary expansion card that costs only a few hundred dollars less than
the whole new Cisco switch you referenced in your reply to my brother, that
just doesn't make sense. (That'e especially true because based on your
email address you may be able to get a significant educational discount on
Cisco.)


> If you do get a switch that supports 10GBase-T, you should definitely
> > consider the Intel X540. The vast majority of reports that I have seen
> say
> > that it works great. (There was one report I found on a forum claiming
> > performance issues, but others on the same thread said it worked fine for
> > them.)
>
>
> Thanks, that's very good to know.
>
>
> > There are also many dual-port SFP+ cards out there (such as the Intel
> X520)
> > that are not too expensive and support lots of different types of SFP+
> > connectors. Although Intel does not make a 10GBase-T SFP+ itself, there
> are
> > third parties that make it. You would use one of those to connect to the
> > 10GbE feed into the rack and then a regular fiber SFP (or the option
> listed
> > below) to connect to the switch.
>
>
> See below for queries/concerns about obtaining a 10GBase-T SFP+
> transceiver.
>
>
> > To connect the pfSense to the switch, I would probably use a
> Direct-Attach
> > cable (DAC) instead of fiber or Ethernet. Approved Optics
> > <http://approvedoptics.com/> is a company that makes many OEM network
> > connectors under contract and they also make their own versions of them
> at
> > significantly reduced prices. Their DAC Finder
> > <http://approvedoptics.com/dac-finder/> tool lets you order a cable that
> > has SFP+ ends for different manufacturers (for example, an Intel end for
> > your pfSense and an HP end for your switch). There's no need to worry
> about
> > fiber or CAT7A Ethernet cables; just plug the cable in (taking care to
> make
> > sure it is oriented correctly) and that's it.
>
>
> Again, many thanks for the Approved Optics link.  That will be very useful.
>
> I don't have any practical SFP+ experience, so maybe you or someone else
> can verify whether I am understanding this correctly: the Direct-Attach
> cable basically encapsulates a transceiver at each end with an appropriate
> cable connecting them, all in one unit?
>

Yes, that us correct. Direct-Attach Cables are usually "twin-ax" cables,
similar to "coax" used for TV and Cable Internet, but with two internal
connectors. The cable is permanently connected to two SFP+ transceivers,
which can be either "Active" or "Passive", meaning that they either have
processing hardware inside the ends or that they pass the signal through
with no changes respectively. Some manufacturers refer to this cable as
"10GSFP+Cu" and others refer to it as "10GBASE-CR".

The major benefit of DA over 10GBase-T is that the Base-T transceiver
latency can be 15-25 times higher than the passive DAC (because the passive
DAC does no processing, while the Base-T needs to process the signal). DAC
also draws almost no measurable power, but the Base-T transceiver hardware
usually consumes 4-8 additional watts. (In your case, with only one runs,
the additional power draw is likely not going to even be noticed, but it
still bears mentioning.)

All that said about the benefits of Direct-Attach, see below for the other
side of the coin.

> Since you have a limited budget, I really recommend going the
> > direct-attached route. They are so much cheaper and more resilient than
> > fiber, and switches with SFP+ slots are often much cheaper than switches
> > with 10GbE. For example, you can get a Uniquiti EdgeSwitch with 48 Gb
> ports
> > and 2 SFP+ ports for just around $400. These are the switches I have used
> > in many of our limited-budget installations in the past (including in a
> > University setting like yours seems to be from your email address) and
> they
> > perform well. (Note that Approved Optics does not have official Ubiquiti
> > cables, but many on the Ubiquiti forums report that it works with Cisco
> and
> > other brand cables as long as they are 2 meters or shorter. In a single
> > rack, that should not be an issue.)
>
>
> My main issue with going the SFP+ route is that my rack uplink port is
> still 10GBASE-T and so I'd need to find a 10GBASE-T transceiver for the
> pfSense 10 GbE NIC and these seem difficult to find or they are 3rd party
> or they are expensive themselves (e.g., $200--$300+).  I've also heard
> there are thermal issues with those transceivers as there's not much
> opportunity to build in the requisite heat sinks that 10GBASE-T appears to
> need.  (I've noticed 10GBASE-T NICs have pretty hefty heatsinks on them.)
> Besides that, I've not been able to find a 10GBASE-T transceiver for
> Chelsio NICs and only 3rd party ones for Intel---e.g., by some company
> called 10Gtek.
>
> Does anyone have any advice/experience to share regarding 10GBASE-T
> transceivers?
>

I don't personally have any of the Intel-compatible 10GBASE-T transceivers
at the moment, but I have seen reports online that the 10Gtek ones are
reliable. (In theory, any SFP+ module that conforms to the official specs
should inter-operate with any other, but, as they say, "that's a nice
theory".)
As far as the heat distribution, that really should be picked up and
handled by the network card and the server's cooling system.


However, based on your response to my brother's email about being able to
budget the Cisco switch with 10GBASE-T, I suggest that you probably
couldn't go wrong with that simply because it's the solution with fewest
number of parts. I would still consider the Intel card over the Chelsio
card if you're really trying to work within a small budget, but you should
go with whatever you think is best for you.


> Thanks again for the info.
>
> Cheers,
>
> Paul.
>


--
Moshe Katz
-- mo...@ymkatz.net
-- +1(301)867-3732
_______________________________________________
pfSense mailing list
https://lists.pfsense.org/mailman/listinfo/list
Support the project with Gold! https://pfsense.org/gold

Reply via email to