2 ms·
Yes, for "just above 1G, and must reuse existing copper wiring, and both sides take PCIE", the card from the article will do. But it's a pretty narrow use case.
by Ao7bei3s 7y ago
Yes, for "just above 1G, and must reuse existing copper wiring, and both sides take PCIE", the card from the article will do. But it's a pretty narrow use case.
I agree with the notion of using fiber. I would certainly stay away from hardwired copper NIC's for anything except 1G, especially rarity ones like the one in the article. Rather, get 10G NIC's with SFP+ cages. Doesn't have to be a pricey Intel card, the cheap Melanox ones are acceptable for most users.
I disagree about OM4. If you're doing new fiber runs, singlemode (OS2 fiber) is the way to go. It's not really expensive anymore. But yeah, OM4 rather than OM3 for all fiber runs (for back to back patches, it doesn't matter ofc).
(Also, since we're talking about fiber networks... my personal lessons learned are: buy reflashable transceivers, especially when that's the difference between having DDM (Digital Diagnostic Monitoring) on all long runs and running it blindly. Standardize on a connector, hopefully LC. Be consistent which side TX and which is RX. Keep a 100-pack of dust caps around. And clean your fibers before inserting, especially patch fibers that get thrown around (they can look pretty bad on the fiberscope). Don't bother with Direct Attach Cables. Stay away from NICs with fixed 10G-Cu ports. Visual fault locators are cheap. BiDi SFP's are wonderful in one-fiber-run-broke or need-another-link-now type emergencies, so keep a few in reserve. And lastly, since all of this costs money, it's easier to beg for reimbursement than to ask for a purchase order.)
- angry_octet 7y agoHmm, lots more transceivers support multimode (not just Ethernet, serial/video etc) and a lot cheaper too. Direct attach is the only way to go for in-rack, both cost and cooling effects. Probably not the best in an office environment though. In the desktop, QNAP recently released 8/12 port SFP+/T switches starting at US$500, which is the sort of pricing that will push this tech into small business. Probably buggy as hell at saturation and no features, but who cares, these are for doing occasional file transfers instead of passing around USB3 SSD drives. https://nascompares.com/2018/04/12/unboxing-the-qsw-1208-8c-qnap-10gbe-switch-its-finally-here/ https://nascompares.com/2018/04/12/unboxing-the-qsw-1208-8c-...
- Ao7bei3s 7y agoI agree MM is more widespread, but I thought we were talking about new cabling (and I took that to include the transceivers). (Serial over fiber? Nice, didn't know that existed.) Pricing-wise, what I meant is: $18: 10G-SR (MM): https://www.fs.com/products/11552.html https://www.fs.com/products/11552.html $24: 10G-LR (SM): https://www.fs.com/products/11555.html https://www.fs.com/products/11555.html Compared to the cost of actually running the fiber through the building, that's not too bad for trunks. As I said, I don't mind multi-mode for the last hop, which is where price is a larger concern due to quantity. And in organizations where OEM pricing is used as baseline, relative to that a resourceful admin can go to single mode and still come out ahead (it's not a fair comparison, but reality in some places). Sure, things are probably different in an actual DC. Direct attach is certainly cheaper (at first), but it can be a hassle compat wise, and I've never had quite the right length available. I can't speak to cooling, that's never been my department.
- angry_octet 7y agoWow those SR SFP prices have fallen. Every SFP adds about 1W, but 10Gb-T ports are particularly thirsty, they were 5W when first introduced, still in that ballpark. It all adds up. Another reason to prefer fiber/DA is 10Gb-T has shit latency, it has a modem doing QAM, hence the power consumption. Probably only relevant for HPC or HFT, but just inelegant.