Any cost efficient methods to get greater than 1Gbps transfers

[anodos]

Maybe if you beg the FreeNAS devs nicely, they will update Samba to 4.4.0 in the next FreeNAS build :)

Samba 4.4.0 supports SMB 3.0 multi-channel, so you can use multiple gigabit ports to transfer data at the aggregated speed of the number of gigabit ports you have connected.

I'm probably just old and cynical, but I wouldn't want to run a Samba x.x.0 release. Note also that the samba release notes say:

Samba 4.4.0 adds *experimental* support for SMB3 Multi-Channel... <snip>
CAVEAT: While this should be working without problems mostly,
there are still corner cases in the treatment of channel failures
that may result in DATA CORRUPTION when these race conditions hit.
It is hence

NOT RECOMMENDED TO USE MULTI-CHANNEL IN PRODUCTION

The all-caps is actually in the release notes. tl;dr, Samba SMB3 multi-channel will kick you in the balls and take your lunch money.

 

[Ericloewe]

I'm probably just old and cynical, but I wouldn't want to run a Samba x.x.0 release. Note also that the samba release notes say:


The all-caps is actually in the release notes. tl;dr, Samba SMB3 multi-channel will kick you in the balls and take your lunch money.

Wow, if the Samba project, of all groups, feels that something is in a dubious working condition (IN ALL CAPS), that's a warning I'd rather heed.

 

[Black Ninja]

It looks like it is some sort of SMT drum core coil, or inductor. As long as the coil itself isn't damaged, it's probably still functional. The surrounding material, since it is clearly a little brittle, appears to be ferrite material. The loss of this isn't a real great thing, but probably isn't catastrophic. I don't think you're going to "know" unless some Intel engineer in here and confirms how vital it is to have that operating in tip-top condition.

Just want to say Thanks

 

[Black Ninja]

Looks like a filtering inductor for a power rail as already suggested (the ceramic caps also confirm this) so it's not critical at all if the ferrite core miss 2 or 3 mm³ on it's end.

I'd buy the card but ask for a small discount because after all it's a damage so the card value isn't the same as the one of a perfect condition card.

Just want to say Thanks

 

[cyberjock]

I wouldn't expect the iXsystems devs to upgrade Samba for this. They used to upgrade it regularly and avoid upgrading except during large upgrade changes (9.3.1 to 9.10) and when there is a security risk because there were too many issues developing due to nuance changes in Samba when it was being upgraded with every release. So they're averse to upgrading unless/until it is necessary. Considering its not recommended for production they aren't going to be interested in adding it. If it were fully baked and reliable, you might have an argument.

HTH

 

[SirMaster]

I'm probably just old and cynical, but I wouldn't want to run a Samba x.x.0 release. Note also that the samba release notes say:


The all-caps is actually in the release notes. tl;dr, Samba SMB3 multi-channel will kick you in the balls and take your lunch money.

Certainly a good disclaimer to add, thanks! FreeNAS 9.10 went to Samba 4.3.4, so what I mean't was perhaps after 4.4.0 gets a few point releases, they may upgrade it in a future release.

Also just because they would add 4.4.x doesn't mean a user *has* to enable and use the multichannel feature. The normal functions would still normally be the same stable functions.

FWIW I've been using the multichannel support in Samba on my home Linux servers from Linux to Linux and Linux to Windows and so far it's been working well. I think it's a pretty welcomed feature in general as most people do have multiple gigabit ports on their machines.

Of course I look forward to them stabilizing the feature now that it's out for everyone to use. That's the first crucial step toward stability and adoption.

 

[Black Ninja]

Dunno about the RPS-720, but let me tell you, the RPS-600 is a watt-burning screamer. I wanna say I caught it jacking up the average power utilization of a 5324 from about 60W without RPS to more like 80W. Which I think may have been per switch, so it might require some substantial power budget and some hearing protection too. But it's been like 10 years since we put those into production so who knows whether my memory is hosed.

I just tossed an MPS-1000 (POE RPS for the Dell 7K's) on the bench and measured it without being attached to a switch and it was 10 watts.

So anyways if it turns out to be a heat or watt burn issue, bearing in mind I don't recall what the inside of the 720 looks like, the inside of the 600 is pretty straightforward:

View attachment 11137

Borrowed from the Internet 'cuz I don't have one to pull apart right now. But basically it's just four Delta PSU's. So whatcha do is you open up the can and disconnect three of them. The AC is the white and black wires at the front conspicuously wired up to the IEC inlet. Then every month you mail the $5.59 you saved in electricity to me, put "beer fund" in the memo, and make me slightly less grinchy.

Very good idea. I'll see if I can disable 3 of them. Little increase in watts is expected when having additional PSU, but don't want to be to much. Switch by itself is around 40W. Power consumption is very very important factor. People don't understand how much these add up to the bill when multiple machines are used.

I should start this fund you mention.

 

[Black Ninja]

They're sitting on some pretty massive traces, which suggests power delivery. It's hard to tell exactly what it's doing from the small area pictured. Inrush limiting doesn't make much sense on a PCI-e card, but maybe some DC-DC conversion on the other side of the board? There's not much cutting-edge stuff that runs off 3.3V these days.

Thank you

 

[Black Ninja]

They are very cost efficient method get 10Gb that's the best way to get faster than 1Gb speed.And they are all here in this great forum. I currently have more 10Gb gear I can use.

P.S. Ones you go fiber you won't go back , and you shouldn't(jgreco warned us about it) . People are afraid if they never use it before, but it's really not that hard and it is worth it. Read the 10Gb primer and don't waste money on keep insisting on making 10Gb work on copper cause you already have the cat6/A wires.That's just my advice.

Black Ninja Support this message.

 

[jgreco]

Oh geez dude you're a hoot. While the current environment heavily favors fiber for 10GbE, I think we have to recognize that sooner or later copper is likely to be the winner. I am perfectly fine with pointing people at fiber as long as they understand that "little caveat." Fiber is all kinds of awesome as long as you're not paying retail, new gear prices.

These Dell N4032F's we have here list for like $12,000 ... each. I can tell you this, I didn't pay that. Since I ran 50/125 instead of the normal 62.5/125 back when we did the original 1G fiber, even though it isn't OM3 rated, works great.

 

[Black Ninja]

No caveats with fiber. It's just more you know about it more you'll love it. Not just latency, or low power , or longer distance, etc. A lot of people forget or don't know that fiber is not affected by Electro-magnetic interference - You can run your fiber along with power cables , around electricity without worried about this problem, which becomes even more of a problem when running 10Gb on copper.

You said that "copper is likely to be the winner" in the 10Gb primer and I wonder since then, why ? Fiber is superior and I don't think all fiber infrastructure(remember the map you showed me) underground will be replaced with copper, I am sure you agree on that. Then if it's better for a backbone of internet, why won't find it's way to every house in US. I kind of actually know the answer to this but don't like it.

They don't need to buy Dell N4032F. They can buy XG2000R from ebay for $400 and have 20x10Gb ports of awesomeness.:)

The 62.5/125 should be OM2 that's why can't be rated as OM3. Is like the cat6 of fiber - can do 10Gb at it's limits on shorter runs. From what I understand, but can't guaranteed I am correct.

 

[anodos]

I wouldn't expect the iXsystems devs to upgrade Samba for this. They used to upgrade it regularly and avoid upgrading except during large upgrade changes (9.3.1 to 9.10) and when there is a security risk because there were too many issues developing due to nuance changes in Samba when it was being upgraded with every release. So they're averse to upgrading unless/until it is necessary. Considering its not recommended for production they aren't going to be interested in adding it. If it were fully baked and reliable, you might have an argument.

HTH

https://bugs.freenas.org/issues/14393

Maybe we'll get a 4.4.1 in 9.10 at some point. :D

 

[jgreco]

You can run your fiber along with power cables

I like dimming my fluorescent lights by wrapping excess lengths of fiber around them until they're sufficiently obscured.

You said that "copper is likely to be the winner" in the 10Gb primer and I wonder since then, why ?

Easy.

1) Fiber involves SFP+ modules (or Xenpak, or XFP, or whatever). These are an additional component, an additional place for problems to crop up. The modules are expensive when purchased new, and might be expensive if purchased used. The Netgear AXM761 lists for $750 and can be bought wholesale for about $250. You need a bunch. Or DAC/twinax cables. List price $120, a third of that retail. Very expensive. And they may be vendor-locked, so if you have Netgear switches and Intel cards, you may need a spare of *both* kinds.

2) Fiber involves fiber. Fiber's hard to work with. It can be damaged more easily than Cat6a, though it offers the potential for higher density.

3) Copper's familiar. It can be field terminated by a monkey with a $20 crimper, and tested for basic wiring correctness with a $10 tester. Spools of cable are cheap. Bags of crimps are cheap.

Fiber is superior and I don't think all fiber infrastructure(remember the map you showed me) underground will be replaced with copper, I am sure you agree on that.

Certainly. For non-local transmission, fiber knocks it out of the ballpark. The fact that I can buy a 10Gbase-ER SFP+ for my switches and run pure ethernet up to 40 kilometers (about 24 miles) is amazing. It'll set me back $5K at each end, but hey...

Then if it's better for a backbone of internet, why won't find it's way to every house in US. I kind of actually know the answer to this but don't like it.

Because it is generally harder to work with. Compare the ease of doing field terminations for RG6 cables... $43 for tools plus consumables for 5 cables. The stuff for telco is even cheaper.. $40. The stuff to do a quality job of terminating fiber in the field? A lot more expensive. And requires training.

See for example http://www.cablinginstall.com/artic...n-fiber-optic-cabling-termination-styles.html

which seems to be a moderately reasonable discussion of the topic. The easiest of these is probably something like the Corning Unicam stuff, but the tool kit alone will set you back $750 and the consumables are probably $30 per full duplex end.

So when the cable/telco truck pulls up to your house and Gunther walks in with his toolbelt and needing crack spackle, do you think he'll have better success with the fiber or the copper?

 

[Ericloewe]

which seems to be a moderately reasonable discussion of the topic. The easiest of these is probably something like the Corning Unicam stuff, but the tool kit alone will set you back $750 and the consumables are probably $30 per full duplex end.

So when the cable/telco truck pulls up to your house and Gunther walks in with his toolbelt and needing crack spackle, do you think he'll have better success with the fiber or the copper?

The el-cheapo fusion splicers seem to start around 1.5k around here. And FTTH ISPs don't trust their field technicians/trained monkeys with those, they all carry around fancier ones that I'd estimate at closer to 7k.

 

[Black Ninja]

I like dimming my fluorescent lights by wrapping excess lengths of fiber around them until they're sufficiently obscured.

Electromagnetism is real. If you don't believe me ask friends, like Faraday for example.:) Interference is the main reason why we can't get the cat5 to work for 10Gb over long distance , not the connectivity. And the reason why theoretical speed is so far of actual.

Copper's familiar. It can be field terminated by a monkey with a $20 crimper, and tested for basic wiring correctness with a $10 tester. Spools of cable are cheap. Bags of crimps are cheap

The monkey can do it. I agree with you. With cheap tools, and cheap cable(I always use high quality cat6 by the way) and the results will be cheap too. Because to test what actually the monkey did is not cheap at all. I don't have this equipment and only use my basic connectivity tested and best skills to terminate and crimp the best way I can.
I've seen tests how connector crimped improper can cause the connectivity issues, didn't even look that bad. Good luck for ISCSI users with that cable.

And about Gunther. I wouldn't trust him to make make connections anyways(I mean copper I still would use premade fiber). I've been Gunther at some time in my life , I wouldn't trust another one even more.

P.S. Gunther's boss is the one to blame mainly for this pathetic job done by Gunther.

 

[Bidule0hm]

Electromagnetism is real. If you don't believe me ask friends, like Faraday for example.

Actually light can also be affected by EM: https://www.youtube.com/watch?v=XhU-nNiAgtI :D

 

[Black Ninja]

Actually light can also be affected by EM: https://www.youtube.com/watch?v=XhU-nNiAgtI :D

Oh Come on. I was expecting at least you to back me up on this one.

 

[Bidule0hm]

You quoted Faraday first... :D

But yeah, in practice fiber is basically immune to EM but copper isn't ;)

 

[Black Ninja]

You quoted Faraday first... :D

I thought you like him.:D:D:D

 

[jgreco]

I thought you like him.:D:D:D

In heavy RF environments we used to have to run network cabling in conduit. Faraday cage *that*.