Need more than 1Gb/s LAN - How can I get there?
-
And be careful. SFP+ cards will still need transceiver modules.
I figured they needed something like that from the shops I saw bundling the cards with transceivers. I think I'll stick with Ethernet for simplicity sake.
-
I'm holding my breath waiting for someone to list some CHEAP 10gb ethernet switches and NICs…
Turning blue....
(-:
-
I figure I'll have to shell out quite a bit. But I'm willing to make that investment. My motto is, you buy cheap you buy twice. So if I have to pay a bit for quality I'll do it as long as I get the right stuff from the start.
-
Well then - You should be happy to know that I've never seen any cheap 10gb Ethernet equipment.
So, you are all set.But yeah - I do think if you need those speeds its the way to go.
-
You say you have a workstation and a server, would it be feasible to forego the switch for now and direct connect?
I don't know if that would work but I'm dying to find out. I'd love to step into 10 gbps for 3 hosts. Contemplating dual port nics in each host and going point to point.
-
Yes I could definitely do point to point and forego the switch. It's something I'm considering to lower costs actually.
-
Here is my situation, I have a desktop and a server. The server contains all my storage, 1Gb (about 112MB/s in my environment) is just not cutting it. I'd really like 400MB/s or more.
I was thinking about using Link Aggregation to bond together four Ethernet 1Gb, but I'm not sure if doing this actually gives me 4x1Gb combined, giving me basically 4Gb network speeds or simply takes a single 1Gb link and splits it across multiple ports/cables so if I had 4 ports I'd get 250Mb/s through each cable culminating in 1Gb max.
I just want to make sure there is no confusion between Gigabit (Gb) vs GigaByte (GB). A 1 Gigabit network has a max speed of 125 MegaBytes per second. According to what you have said your network is running at 896 Megabit (112 MegaBytes x 8 bits in a Byte).
As for Link Aggregation, it will give you 4 sessions at 1 Gb, and not 1 session at 4 Gb, even if it states 4 Gb on the bonded interface. In my experience and testing traffic will only move across a single interface, and the traffic does not split across several of them.
With that said, I setup a 10 Gb network at my work for production servers and for VM storage. The speeds I have seen average about 75 MB/s with occasional spikes to 400 MB/s, and that is with using SSDs and RAID. Even our 1 Gb iSCSI and Fiber iSCSI devices hardly see their max speeds when being used.
Obviously without knowing your environment and requirements, it is hard to say if going 10 Gb would be worthwhile or a waste of time and resources. When you do go 10 Gb though you end up finding new bottlenecks like bus speeds, cabling issues, drive speeds, and overhead. As an example, SATA 2 has a max of 300 MB/s and really good SSDs have a max of 600 MB/s, and that's without any bottlenecks.
-
That's correct. My network right now has a nice sustained transfer speed of 112MB/s aka 1Gb/s with overhead.
For example if I were to transfer a 30GB file right now from my server to my desktop it transfers the entire way (SMB 3.0 on both sides) at 112MB/s with no dips. My server is running 9 x 4TB Hitachi disks in RAID6 on a high end hardware card. The desktop is using SSD's with 500MB/s write rates, high end SSD's
What I'm looking for is performance around 4Gb/s or lets say around 400MB/s (3.2Gb/s). Anything above that would be nice but not needed.
I think what I'm going to do is buy two Intel 10Gb cards and link the Desktop to the Server directly and see what that yields.
-
It'll be interesting to see what happens then with the 10 Gb NICs. If that is not fast enough, I would look at changing from RAID 6 to RAID 10 if possible. I know with my backup server that change made the difference from backups done at noon to backups done when I got to work. Obviously, you lose a bit in capacity, but you no longer have the parity overhead that RAID 5 and 6 have.
-
I get over 600MB/s sustained with my current setup in RAID6. Not worried about that not being able to cope at all :)
-
FYI, SMB3 is capable of using multi-path is nearly every way possible. Be it multiple interfaces, each with different IP addresses or a single interface that is bonded, it will attempt to make multiple connections. One of the things specifically being addressed was bonded interfaces that use hash based teaming.
In theory, you should be able to get near 4Gb/s with a teamed 4x1Gb link with SMB3, assuming the best case of large file transfers or even block-device access. SMB3 is comparible or better than iSCSI.
I'm not sure if this is out of box or you need to configure it, but I remember reading about these major changes.
-
And be careful. SFP+ cards will still need transceiver modules.
only if using fiber, if using twinax the transceiver is part of the cable (it also uses less power and has lower latency)
If you want to use cat6a/7 be aware that it will have much higher cable loss and because of this you cant do FCoEwhile 10GB Ethernet is ideal it is still so expensive and out of the reach of most home uses (this will change). If you have the spare switch ports test out multiple 1Gb links Server 2012R2 and SMB 3 supports multipathing so where you were limited to the throughput of a single link in a team previously now you should be able to take advantage of all the active links
-
Sure. Plenty of 40Gb twinax out there.
-
This link might help you out:
https://forums.servethehome.com/index.php?threads/basic-10gb-setup-for-two-pcs.2105/It leans towards an inexpensive point to point QDR infiniband setup. Two nics can be had for around $100 ish and connect x-2 support RDMA with the OS(s) you're using. Could save you a bit of money over going 10gbe and it's quite a bit quicker.
