The company I work for uses wireless differently than most people, and we've come across an issue and need some help. We upload massive amounts of data from multiple Groove devices to a single (or dual) SXT access points. This was my first time testing with CAPsMAN, so maybe I set something up wrong, but I don't know what to check as everything seemed to be correct, just slow. Any insight or ideas to check? With everything being gigabit and no resource issues on the server, I'm not sure why we're hitting such low limits.
All devices set to 5GHz-only-N.
All devices running RouterOS 6.37 with latest firmware.
Test scenario 1:
8x devices uploading 3-5GB each from a Groove 52HPn to a single SXT SA5 with a gigabit connection to a server.
We started uploads in 2 minute intervals and noticed that the SXT peaked around 90Mbps from 3 clients all the way to up to 8 clients.
Test scenario 2:
8x devices uploading 3-5GB each from a Groove 52HPn to a two SXT SA5 connected to an RB450G running CAPsMAN and connected to server.
We started uploads in 2 minute intervals and noticed that the combined upload peaked between 60-65Mbps with and without load balancing.
Re: Wireless Upload Speeds
How is the CPU load on the Mktik devices once the upload is running?
Re: Wireless Upload Speeds
These are all based on the 5 minute average graphs.
Test Scenario 1;
AP CPU peaked at 33%, Memory peaked at 31%
Grooves CPU peaked at 31-32%, memory peaked at 32-33%
Test Scenario 2:
RB450G CPU peaked at 43%, Memory peaked at 30%
Both APs CPU peaked at 26%, Memory peaked at 34%
Test Scenario 1;
AP CPU peaked at 33%, Memory peaked at 31%
Grooves CPU peaked at 31-32%, memory peaked at 32-33%
Test Scenario 2:
RB450G CPU peaked at 43%, Memory peaked at 30%
Both APs CPU peaked at 26%, Memory peaked at 34%
Re: Wireless Upload Speeds
what channel width are you using ?
what modulation rates are the clients connected at ?
what modulation rates are the clients connected at ?
Re: Wireless Upload Speeds
Channel: 5805/20-Ce/an(30dBm)
TX Rate: 150Mbps-40MHz/1S/SGI
RX Rate: 150Mbps-40MHz/1S/SGI
CCQ between 90-94%
Not sure if that's what you mean by modulation rate. If not, let me know.
TX Rate: 150Mbps-40MHz/1S/SGI
RX Rate: 150Mbps-40MHz/1S/SGI
CCQ between 90-94%
Not sure if that's what you mean by modulation rate. If not, let me know.
Re: Wireless Upload Speeds
Given your modulation rates, it sounds like you're getting about the capacity one could expect.
First: The modulation rate is the data rate of the wireless link itself, NOT the data rate you can expect to see with payload. 802.11 has a very high amount of overhead in the protocol, and my personal rule of thumb has always been that you can expect to see roughly half of the modulated rate as payload rate... (this is an overly-conservative estimate, but it served me well back when I had to deal with wireless services on a regular basis).
Second: WiFi is half-duplex. Simply put, the 150Mbps of modulated throughput must be shared between upload and download streams. So you could get 100% up / 0% down, 60%up / 40% down, 30%up/70% down, etc. This applies to both the actual throughput with overhead and the payload throughput rate... meaning that if you can expect ~80Mbps of payload throughput, then that 80 gets shared between up/down.
Third: TCP generates ACK packets as a part of its functionality, and this traffic can become non-trivial when a lot of throughput is being transferred. Basically, as the upload speed goes up, so too will the downstream traffic of ACK packets from the server.
Fourth: Wireless doesn't actually share bandwidth in throughput rates - it shares in time slices. This means that you won't get as much throughput if there are lots of small packets (like ACK packets) going through your network. It would be like contracting a cargo ship to send one car. The ship is going to charge the same amount to sail whether you put one car on it or fill it to capacity. What this means is that the ACKs are taking away more capacity than their size would seem to do. (this is actually just normal, but it's a factor)
Fifth: I don't have the data to determine this from your situation, but I wanted to share this as another factor - given the time sharing situation, it means that any wireless client with a poor modulation rate can diminish the capacity of an entire AP. Whenever the slower client gets its turn, it takes longer to transmit the same amount of data than a faster client would do. So a slow client is taking up the same amount of time but transferring less data -> max throughput in payload gets decreased.
Sixth: The capacity of the AP is shared between all clients. (this seems obvious but I wanted to enumerate it as a consideration). Even though each client may be modulated at 150Mbps, they must all share the same time slices. It's not quite as simple as saying "the capacity of the AP is the same as the slowest client" or saying "the capacity is the same as the fastest client." It just depends on which clients are active and which aren't, and how much capacity they're attempting to use.
All this is to say that getting ~90Mbps of throughput on a WAP that's modulating 150Mbps links seems fair to me. I see these are 5Ghz band radios. If your RF environment is relatively clear, you could try bumping up the channel width to get 300Mbps mod rates.
First: The modulation rate is the data rate of the wireless link itself, NOT the data rate you can expect to see with payload. 802.11 has a very high amount of overhead in the protocol, and my personal rule of thumb has always been that you can expect to see roughly half of the modulated rate as payload rate... (this is an overly-conservative estimate, but it served me well back when I had to deal with wireless services on a regular basis).
Second: WiFi is half-duplex. Simply put, the 150Mbps of modulated throughput must be shared between upload and download streams. So you could get 100% up / 0% down, 60%up / 40% down, 30%up/70% down, etc. This applies to both the actual throughput with overhead and the payload throughput rate... meaning that if you can expect ~80Mbps of payload throughput, then that 80 gets shared between up/down.
Third: TCP generates ACK packets as a part of its functionality, and this traffic can become non-trivial when a lot of throughput is being transferred. Basically, as the upload speed goes up, so too will the downstream traffic of ACK packets from the server.
Fourth: Wireless doesn't actually share bandwidth in throughput rates - it shares in time slices. This means that you won't get as much throughput if there are lots of small packets (like ACK packets) going through your network. It would be like contracting a cargo ship to send one car. The ship is going to charge the same amount to sail whether you put one car on it or fill it to capacity. What this means is that the ACKs are taking away more capacity than their size would seem to do. (this is actually just normal, but it's a factor)
Fifth: I don't have the data to determine this from your situation, but I wanted to share this as another factor - given the time sharing situation, it means that any wireless client with a poor modulation rate can diminish the capacity of an entire AP. Whenever the slower client gets its turn, it takes longer to transmit the same amount of data than a faster client would do. So a slow client is taking up the same amount of time but transferring less data -> max throughput in payload gets decreased.
Sixth: The capacity of the AP is shared between all clients. (this seems obvious but I wanted to enumerate it as a consideration). Even though each client may be modulated at 150Mbps, they must all share the same time slices. It's not quite as simple as saying "the capacity of the AP is the same as the slowest client" or saying "the capacity is the same as the fastest client." It just depends on which clients are active and which aren't, and how much capacity they're attempting to use.
All this is to say that getting ~90Mbps of throughput on a WAP that's modulating 150Mbps links seems fair to me. I see these are 5Ghz band radios. If your RF environment is relatively clear, you could try bumping up the channel width to get 300Mbps mod rates.
Re: Wireless Upload Speeds
ZeroByte, that helps alot. I've done networking for a few years now, but I'm getting more into the wireless side and it all seems more magic than science so far. Your explanation makes sense and gives me more insight into what is actually happening. I'll see if I can get the 300Mbps rates running at least in my test environment even though I know I won't be able to use it in production.
Re: Wireless Upload Speeds
I think the first key to understanding wireless is knowing the difference between modulated rate and signal strength.
They're related in some ways, but they're not dependent on each other at all.
Mod rate can only go up with a clean signal - i.e. a high signal/noise ratio.
You could listen to someone talking really fast at a whisper in a very quiet room with no echoes...
If the room is still quiet but very "echo-ey" they would have to slow down so that you could make out the individual words.
If the room is noisy, they can shout, and you can hear their voice, but if it's getting mixed with the other noise, they still have to slow down.
Another way is to think of vision- if I made a sign that was very low contrast - say dark green letters on a very-slightly-lighter green background - You couldn't make out very tiny print, no matter how well-lit. You could tell there's writing, but it would be difficult to distinguish the letters. It wouldn't matter how brightly the sign is lit (signal strength) So you have to waste more space on the paper by printing larger, thus there is less information per page.
If it's black on white, then you can distinguish it much more easily and can read smaller type in dimer light....
If you'd like to read up on why this is, go read about QAM modulation - the Wikipedia article is excellent.
(the QAM rate like QAM-4, QAM-16, QAM-64, etc... would be analogous to the font size above)
They're related in some ways, but they're not dependent on each other at all.
Mod rate can only go up with a clean signal - i.e. a high signal/noise ratio.
You could listen to someone talking really fast at a whisper in a very quiet room with no echoes...
If the room is still quiet but very "echo-ey" they would have to slow down so that you could make out the individual words.
If the room is noisy, they can shout, and you can hear their voice, but if it's getting mixed with the other noise, they still have to slow down.
Another way is to think of vision- if I made a sign that was very low contrast - say dark green letters on a very-slightly-lighter green background - You couldn't make out very tiny print, no matter how well-lit. You could tell there's writing, but it would be difficult to distinguish the letters. It wouldn't matter how brightly the sign is lit (signal strength) So you have to waste more space on the paper by printing larger, thus there is less information per page.
If it's black on white, then you can distinguish it much more easily and can read smaller type in dimer light....
If you'd like to read up on why this is, go read about QAM modulation - the Wikipedia article is excellent.
(the QAM rate like QAM-4, QAM-16, QAM-64, etc... would be analogous to the font size above)