Infiniband raw speed is not the same as 10GBE. Infiniband 4x 2.5 Gbps per lane with 8/10b encoding datarate = 2 Gbps per lane 4 lanes * 2 Gbps/lane = 8 Gbps this is datarate in 1 direction. double if going in both directions.
10GbE 3.125 Gbps per lane with 8/10b encoding datarate = 2.5 Gbps/lane 4 lanes * 2.5 Gbps = 10 Gbps this is datarate in 1 direction. double if going in both directions. I have no interest in Infiniband 4x so haven't and won't research those interconnects. I haven't yet reserached 10GbE CX4 commercial network hardware to see if any provide the power on the appropriate pins for the FO links. I wouldn't be surprised in there very few, or even none as you report, that do so. On Tue, 17 Jun 2008, Jouko Ritakari wrote: > Hi Matt, > > I understand that Infiniband raw speed is the same as 10GBE. > > Most of the physical alyer standards (for example CX4) are compatible too, > but the 10GBE people left the Infiniband 4-fiber and 12-fiber modes out. > You cannot buy 10GBE switches that support these anywhere. Nor can you buy > the transceivers. > > Cheers, > Jouko > > "Life is pretty simple: You do some stuff. Most fails. Some works. You do > more of what works. If it works big, others quickly copy it. Then you do > something else. The trick is to do something else." > > > On Fri, 13 Jun 2008, mdex...@berkeley.edu wrote: > > > Hi Jouko, > > What do you mean by Infiniband only ? > > > > As mentioned before we are using them at 4 lanes * 3.125 Gbps not > > Infiniband's 4 * 2.5 Gbps. > > > > Matt > > > >> Hi Matt, > >> > >> I did some research about the fiber optic links some months ago. > >> > >> It appears that the Fujitsu things are infiniband-only, they cannot be > >> connected to 10 Gbps Ethernet switches directly. Or more precicely, need > >> another such device to connect to a copper port. > >> > >> I did try to convince the roach people to design a footprint for a > >> mainstream fiber optic transceiver, didn't succeed. As far as I know there > >> are about five 10GBE transceiver standards around, three of which have > >> same signals as CX4. So it would have been a zero-cost option, just add > >> some wires in the PCB. > >> > >> Cheers, > >> Jouko > >> > >> "Life is pretty simple: You do some stuff. Most fails. Some works. You do > >> more of what works. If it works big, others quickly copy it. Then you do > >> something else. The trick is to do something else." > >> > >> > >> On Thu, 12 Jun 2008, Matt Dexter wrote: > >> > >>> Hi, > >>> > >>> We are starting to run some tests with one of the Cognitive Radio > >>> group's > >>> Fujitsu o-microGigaCN (o-MGC) fiber optic links. This unit is on > >>> loan and will need to be returned in a bit. > >>> > >>> This consists of: > >>> Part number Serial number description > >>> FPD-010R008-OE 01B-06160045 CX4/FO transceiver, 850nm VCSEL, 3.3V > >>> FPD-010R008-OE 01A-05430042 CX4/FO transceiver, 850nm VCSEL, 3.3V > >>> FOC-CC10004 50/125um MMF 850nm wavelength approx 50ft > >>> > >>> optical interface is MPO connector > >>> electrical interface is microGiGaCN > >>> > >>> http://www.fujitsu.com/us/services/edevices/components/connectors/copper-modules > >>> > >>> These units are reportedly quite expensive; maybe $1K per link > >>> (2*$400+200). > >>> > >>> Shortly I'll work on getting revised information but back in April of > >>> 2005 > >>> I was told: > >>> Bob Thornton is the sales contact for the Fujitsu modules. His contact > >>> information is: > >>> Director of Marketing > >>> Fujitsu Components America, Inc > >>> bthorn...@fcai.fujitsu.com > >>> 408-745-4932 > >>> www.fcai.fujitsu.com > >>> He will be receiving the first production modules in the next several > >>> weeks. He estimates the prices to be: > >>> module $350; > >>> 3m patch cord $90. > >>> Lead time: 2-3 weeks. > >>> > >>> The attachments were provided, back in 2005, by Fujitsu. > >>> I haven't found them online. > >>> > >>> > >>> > >>> Oren is running through some tests now. Additional results will be > >>> distributed later. For now > >>> iBob->iBob seems aok. > >>> > >>> Connecting to the front of the BEE2s doesn't seem safe without > >>> re-machining the lower edge of the BEE2 chassis' front opening. BWRC > >>> folks designed this lower edge to be a "pinch bar" of sorts to help > >>> hold > >>> the CX4 cable assemblies in place. The FO unit becomes taller sooner > >>> than the WLGore CX4 cable assembly for example and thus the angle into > >>> the BEE2's infiniband 4x connectors looks to be too steep. We didn't > >>> try to force it in. > >>> > >>> There is no "pinch bar" of any type in the back of the chassis so there > >>> are no obstructions. Oren is trying those out now. > >>> > >>> iBob-> back of BEE2 OK > >>> back of BEE2 -> back of BEE2 OK > >>> back of BEE2 -> iBob OK > >>> > >>> Over the next few days we plan on testing more extensively. > >>> > >>> Matt > >>> > >>> ------------------------------------------------------------------------ > >>> > >>> Date: Fri, 2 May 2008 11:53:41 -0400 (EDT) > >>> From: John Ford <jf...@nrao.edu> > >>> To: Matt Dexter <mdex...@berkeley.edu> > >>> Cc: David MacMahon <dav...@astro.berkeley.edu>, > >>> casper@lists.berkeley.edu > >>> Subject: Re: [casper] iBOB to iBOB over 15m CX4 > >>> > >>>> On of the non-Casper BWRC groups has a box of the fiber optic cables > >>>> Dave mentions. We got the OK to borrow a couple to exercise them for > >>>> a few days. This is high priority on the to-do list. > >>>> > >>>> Note as we as allowing longer distances the fiber optic cables > >>>> may be easier to route than the copper versions. Shorter radius of > >>>> curvature and smaller diameter. > >>> > >>> Hi Matt. Could you post the vendor or other info on these cables? I'm > >>> interested in using something similar. > >>> > >>> Thanks. > >>> > >>> John > >> > > > > > > >