NETGEAR is aware of a growing number of phone and online scams. To learn how to stay safe click here.
Forum Discussion
Solved
Questions About Ethernet Port Link Aggregation
I've been reading about link aggregation and I think I understand the basics. Linking two ethernet ports will not double the speed, but it will double the bandwidth. So I still have some quesitons:...
Video streaming isn't really that demanding over gigabit ethernet. Even 4K 3-D video won't exceed 100 megabits, and that is only 10% of your gigabit bandwidth. And most 4K is much less.
FullHD BluRay won't exceed ~54 megabits (5% of your gigabit bandwidth).
Most video downloads use much lower bitrates (FullHD mkv files are often around 8-10 megabits).
So for your specific use case, you probably don't need Link Aggregation.
XrayDoc88 wrote:
...Linking two ethernet ports will not double the speed...
That depends on the details. If the device accessing the NAS is also using link aggregation, there are some bonding modes that would double the speed.
XrayDoc88 wrote:
2. Since the bandwidth from the ReadNAS to the switch would be doubled, wouldn't this help with the data flow from the NAS when streaming to two separate computers on the same LAN? I'm thinking of the situation where two people are watching two different movies on two different smart TVs or on two different computers.
It can help in this case. It is not guaranteed to improve the performance, but it shouldn't hurt performance.
If you are using LACP, then the system chooses a specific NIC for each data flow. That's what the hash mode does. If you only have two computers accessing the NAS, there is a 50-50 chance they will end up using the same NIC. When that happens, the performance is the same as not using aggregation.
If you choose a static LAG on the Cisco device, then you'd want to choose the "round-robin" mode in the NAS. The NAS will then alternate the packets (packet 1 goes on NIC-1, packet 2 goes on NIC-2, ...) That can double the speed (if the receiver is also bonded using a static LAG).
But if the receiver isn't bonded, then it can result in lost packets (because the NAS is delivering packets faster than gigabit ethernet can carry them). In this case, It's best to enable ethernet flow control on the receivers (and if you have smart switches, on the switches too).
XrayDoc88 wrote:
3. Since the aggregated link would only be between the ReadyNAS and the switch, should I choose a "static" LAG?
I'd suggest LACP, as that avoids the possibility of the packet loss/queue overflow issue I mentioned above. If you were using aggregation in the NAS clients as well, then I'd consider using a static LAG.
XrayDoc88 wrote:
4. There were lots of other choices for the type of Link Aggregation on the ReadyNAS, including some "hashes" I believe. Which choices would you recommend, or do those choices only apply for dynamic LAGS with LACP enabled?
For static LAG, just use round-robin.
For LACP, either layer 2 or layer 2+3 are reasonable. In practice it doesn't matter much.
Also, the hash mode is only about selecting the NIC for transmission of each packet. The switch won't know what hash the NAS is using, and it often will use a different one. The Netgear switches I use don't allow me to configure the hash on the switch (and I believe they use layer 2).
Video streaming isn't really that demanding over gigabit ethernet. Even 4K 3-D video won't exceed 100 megabits, and that is only 10% of your gigabit bandwidth. And most 4K is much less.
FullHD BluRay won't exceed ~54 megabits (5% of your gigabit bandwidth).
Most video downloads use much lower bitrates (FullHD mkv files are often around 8-10 megabits).
So for your specific use case, you probably don't need Link Aggregation.
XrayDoc88 wrote:
...Linking two ethernet ports will not double the speed...
That depends on the details. If the device accessing the NAS is also using link aggregation, there are some bonding modes that would double the speed.
XrayDoc88 wrote:
2. Since the bandwidth from the ReadNAS to the switch would be doubled, wouldn't this help with the data flow from the NAS when streaming to two separate computers on the same LAN? I'm thinking of the situation where two people are watching two different movies on two different smart TVs or on two different computers.
It can help in this case. It is not guaranteed to improve the performance, but it shouldn't hurt performance.
If you are using LACP, then the system chooses a specific NIC for each data flow. That's what the hash mode does. If you only have two computers accessing the NAS, there is a 50-50 chance they will end up using the same NIC. When that happens, the performance is the same as not using aggregation.
If you choose a static LAG on the Cisco device, then you'd want to choose the "round-robin" mode in the NAS. The NAS will then alternate the packets (packet 1 goes on NIC-1, packet 2 goes on NIC-2, ...) That can double the speed (if the receiver is also bonded using a static LAG).
But if the receiver isn't bonded, then it can result in lost packets (because the NAS is delivering packets faster than gigabit ethernet can carry them). In this case, It's best to enable ethernet flow control on the receivers (and if you have smart switches, on the switches too).
XrayDoc88 wrote:
3. Since the aggregated link would only be between the ReadyNAS and the switch, should I choose a "static" LAG?
I'd suggest LACP, as that avoids the possibility of the packet loss/queue overflow issue I mentioned above. If you were using aggregation in the NAS clients as well, then I'd consider using a static LAG.
XrayDoc88 wrote:
4. There were lots of other choices for the type of Link Aggregation on the ReadyNAS, including some "hashes" I believe. Which choices would you recommend, or do those choices only apply for dynamic LAGS with LACP enabled?
For static LAG, just use round-robin.
For LACP, either layer 2 or layer 2+3 are reasonable. In practice it doesn't matter much.
Also, the hash mode is only about selecting the NIC for transmission of each packet. The switch won't know what hash the NAS is using, and it often will use a different one. The Netgear switches I use don't allow me to configure the hash on the switch (and I believe they use layer 2).
Thank you Marc_V and StephenB. Very helpful information. I am a little dizzy, but I haven't passed out yet. Actually, I think I followed most of what was written. :)
I had no idea about the required bandwidth for video. I thought streaming two 4k movies simultaneously would really tax the NAS and a single Gbps connection, and teaming might help. Plus, I have an unused port on my switch so I thought I might as well give it a try.
So static round robin doesn't sound fool proof if I follow correctly. When streaming one movie, packet 1 and then packet 2 are arriving at the switch over two pipes, faster than a single Gbps connection, but then go out to the Smart TV over a single connection and I could lose packets, correct?
With LACP all packets from the same movie should travel over the same pipe to the switch, correct? That particular movie would not travel any faster than Gbps to the switch?
With LACP, when you start streaming the second movie from the NAS, hopefully it gets assigned to the second bonded pipe and is not competing with the first movie, correct? But there's no guarantee that will happen? I guess LACP should make an "intelligent" choice based on the current data flow traffic that the bonded pair are experiencing?
The instructions article that Marc_V referenced here, is helpful and I assume it will be very similar for the 628X. But my model isn't actually listed at the bottom of those instructions. I'm assuming it's just an older article that hasn't been updated.
Finally, I made an unrelated observation last night. I'm transitioning from an old Windows Home Server machine I built in 2009 to the ReadyNAS. The WHS machine has 7200 rpm hard drives and a Gbps NIC. As I was copying files to the ReadyNAS, the fastest speed I was seeing was about 80 MB/s. I guess this equates to about 640 Mbps? Does that seem reasonable because of hard drive speeds or other network traffic? Or should I be seeing closer to 1000 Mbps?
Thanks again!
HI XrayDoc88
It should be the same for 628X since it's an OS6 system as well.
We encourage you to mark the appropriate reply as the “Accept as Solution” so others can be confident in benefiting from the solution.
The Netgear community looks forward to hearing from you and being a helpful resource in the future!
Regards
XrayDoc88 wrote:
So static round robin doesn't sound fool proof if I follow correctly. When streaming one movie, packet 1 and then packet 2 are arriving at the switch over two pipes, faster than a single Gbps connection, but then go out to the Smart TV over a single connection and I could lose packets, correct?
Correct. One solution is to enable ethernet flow control in all the devices, and that will allow the switch to pause the transmission when it's queues overflow. That more or less eliminates the benefit, since the flow control will pause all traffic coming out of the NAS until the quue drains. But it will prevent the packet loss.
XrayDoc88 wrote:
With LACP all packets from the same movie should travel over the same pipe to the switch, correct? That particular movie would not travel any faster than Gbps to the switch?
Correct. LACP is designed to ensure that each data flow won't exceed one megabit.
XrayDoc88 wrote:
With LACP, when you start streaming the second movie from the NAS, hopefully it gets assigned to the second bonded pipe and is not competing with the first movie, correct? But there's no guarantee that will happen? I guess LACP should make an "intelligent" choice based on the current data flow traffic that the bonded pair are experiencing?
The LACP standard allows for some intelligence, but in practice the hashing algorithm determines what NIC is used. For instance, you can XOR the source and destination MAC address MOD the the number of cables in the bond. That is completely deterministic, and doesn't depend on the amount of traffic on each link.
Link aggregation was originally developed to connect switches, and there was a tacit assumption that there would be a lot of data flows running between the switches. When that's the case, the traffic does end up balanced fairly well, even with simple algorithms.
XrayDoc88 wrote:
Finally, I made an unrelated observation last night. I'm transitioning from an old Windows Home Server machine I built in 2009 to the ReadyNAS. The WHS machine has 7200 rpm hard drives and a Gbps NIC. As I was copying files to the ReadyNAS, the fastest speed I was seeing was about 80 MB/s. I guess this equates to about 640 Mbps? Does that seem reasonable because of hard drive speeds or other network traffic? Or should I be seeing closer to 1000 Mbps?
80 MB/s is about 640 Mbps, yes. There is overhead in the protocol, so you won't every see 1000. Buf 640 is a bit lower than I'd expect. You might try testing with NAStester ( http://www.808.dk/?code-csharp-nas-performance) or map the NAS to a drive letter on the WHS server, and test with CrystalDiskMark.
My main NAS is a 526x (not as fast as your 628x), but I am using 10 gigabit ethernet. When I access it from an application server that also has 10 gigabit, I see transfer speeds well over 300 MB/s. So your NAS is definitely capable of exceeding your network speed for large-file transfers.
NETGEAR Academy
Boost your skills with the Netgear Academy - Get trained, certified and stay ahead with the latest Netgear technology!
Join Us!
