I conducted tests on a 5.7 network with non-advantage products to see what throughput could actually be realized from one SM to another SM. The results were not so great.
I used a shareware product called Bandwidth Meter, which can be downloaded from http://downloads-zdnet.com.com/BWMeter/ … ag=lst-0-5
It has a feature where you can start a timer, perform some type of file transfer or data transmission, and it calculates the transfer speed. The units of speed are in bytes per second, and can also be viewed in KB, MB, GB, and TB.
The test I performed was the following. It was a test from SM to SM on the same cluster. Both sides are behind Linksys NAT routers. I setup the one router to port forward Windows File Sharing ports directly to a Win2k Domain Controller, which had the files that I wished to D/L through the
Canopy network to test the speeds. I forwarded 135-139, and 445. I downloaded a 20MB file from this server to my laptop, which again is also behind a Linksys router. It took approximately 4 min and 40 seconds to download the file. After converting the units from MB per second to Mb per second, the maximum transfer rate was a whopping .9 Mbps. Thats right, not even 1 Meg.
I started thinking of possible variables which could be bottlenecks. The only change I made was connected the SM at my location directly to my laptop, to avoid the processing and NAT traversal that was taking place at the Linksys router at my end. I also disabled my personal firewall, so it did not have to inspect every packet sent and received from my computer. The speed increased, but was still under 1 Meg.
Other possible bottlenecks are the Linksys processesing at the server end, and the amount of memory that is in the server. As I said before, both SMs, and the 2 APs that they are registered to are 5.7Ghz, and non-advantage. The SM at the remote end is under 1 air mile from our tower, has an RSSI of well over 1300+, Jitter of 1-2. The SM at my location is going through some trees with no leafs (leaves?) at 6.5 miles out, but with an RSSI of 980+, Jitter of 2-3. Link tests on both SM’s are 100% uplink and downlink. Link tests from the APs that these SMs are connected to are 100%. Configuration source on both APs are set to SM. I have both SMs set to allow maximum upload and download speeds.
I realize that round trip latency from one SM to another is going to be in the 40-50 ms range since it is non-advantage stuff, but under 1 Meg is just not acceptable.
Just as a side note, thinking that the source of the problem might be the software I was using, I directly connected one laptop to another using a crossover cable, and transferred the same 20MB file from one machine to another. The program reported that the max trasfer rate was 50Mbps. That seems awful slow to me.
Also, I think that Microsoft File Sharing traffic uses UDP as its transport layer protocol, so there is not as much overhead (i.e. ACKs in TCP) as TCP. I did not sniff traffic while I was doing this to see if it was the case.
Any ideas? Has anyone else performed tests like this?
What upload bandwidth do you get if you FTP a file to a server that is on the wan side of the ap? Is it the same or substantially more?
What happens if you change the downlink to 50% to give you more uplink bandwidth?
Jerry Richardson
airCloud Communications
When the “Configuration Source” parameter on the AP is set to SM, the Downlink Percentage setting on the AP doesn’t matter (at least I think so).
I think that might answer your question I read a few minutes ago. If are not using BAM, you will set your AP’s configuration source to either AP or SM. If its set to AP, then the bandwidth for all SMs is controlled by the Downlink Percentage parameter. If it is set to SM, then the SM controls its own bandwidth.
I tried changing the Max Range setting on our APs from a value of 15 to a value of 7 after reading some of the posts about throughput. This increased throughput from .9 Mbps to 1.1 Mbps. Still not acceptable. One of the things that tech support’s “product specialist” told me was that the Receive Window on a Win2000 or WinXP machine is maxed out at 1 Meg, so the fastest a single file will transfer at using Windows File Sharing is 1 Meg. Completely incorrect. Why can I transfer files on a 100 Mbps LAN at close to 100 Mbps if the max is 1 Mbps? Receive Window is a header in a TCP packet. Windows File Sharing uses UDP.
The other thing he recommended was disabling the “Transmit Frame Spreading”. It apparently kills bandwidth by 10%. If the throughput of an SM is roughly 6.2 to 6.8 Mbps, then 10% loss is only going to be 62 - 68 kbps. I am losing way more bandwidth than that.
Any ideas?
Correction - Windows File Sharing uses both TCP and UDP. 135 - 139 and 445. The Receiver Window response still does not make logical sense.
Downlink Data % (set on the AP’s Configuration page) sets downlink % for the sector. It is not affected by the Configuration Source option.
The SM data controlled by Configuration Source includes MIR, VLAN, CIR (hw sched only) and high priority (hw sched only).
If you have the Downlink Data % set to the default of 75% down, that means 25% up, or .25*6.2 or about 1.6 Mbps. Sw sched takes two SMs typically to generate full use of the bandwidth just because of the way it works. So, your 1.1 Mbps is about right for a single SM to SM connection.
Try changing the 75% downlink to 50% and see what you get. You may see in the neighborhood of 1.5 or more Mbps. And if you add another 2 SMs to the AP and run the same test between them simultaneously with running it with your first pair, you should develop the full 3.1 Mbps (half of 6.2) with both links.
I know you were focused on understanding behavior of standard APs, but if you were to test an Advantage AP in the same scenario, you might see 3.5 Mbps because the hardware scheduler gets 7 Mbps aggregate and can develop the full bandwidth to a single SM. And if you could get 2X operation (maybe doubtful with one of your distances) you might see bursts up to 7 Mbps from one SM to the other.
Another thought - latency has a lot of impact on the “feel” of speed if the traffic is acknowledged. All that handshaking goes so much faster with lower latency, that even though the raw speed may be the same, a connection with lower latency will feel much faster.
Please do tell if you get 1.5 Mbps with 50% downlink setting.
Thanks for the info johnh. I conducted a few tests to try to optimize bandwidth. The first thing I tried was changing the Max Range on all AP’s from 15 miles to 7 miles. According to a previous tech support post, if you have it at at 1-14 miles, you use one control slot, and if you max it out at 15 miles, you use another control slot, limiting bandwidth. I first started at 7 miles, and noticed that all Link tests from SMs were affected. Instead of being 100%, they dropped to 95-98%. The only settings for Max Range that provided 100% link tests were 14 and 15. I left it set to 14.
I then disabled Transmit Frame Spreading, and monitored the sector for a period of time to make sure nothing changed. Everything was fine, so I tried a bandwidth test, and I was still around 1.5Mbps. I then disabled encryption. According to tech support, and written documentation, applying encryption has NO effect on throughput. I don’t buy that one. Ciphering and deciphering the bits takes time, it may not be much time, but its still time. Writing a program that simply displays text to the screen is going to take less time to run than a program that ciphers that text, deciphers it, then finally prints it. Tack on a 25ms latency time, and it can add up. In my opinion, you might not be able to notice a change in throughput, but I would think in theory that it exists. I kept encryption turned off. Tests were still around 1.5 Mbps.
Finally, I wanted to see if tech support was correct when they told me that setting the config source to SM causes the Downlink Data Percentage to be obsolete. I switched it to somewhere around 10 - 15%, with my SM bandwidth maxed, and yes I saw that my download speed to the internet decreased significantly. So, this meant that the Downlink % param still has an impact on throughput even those Config Source is set to SM. I changed it to 50% on all AP’s, and a single file download was still at 1.5Mbps. So, I used as much bandwidth as I could at once, downloaded a 29MB file from the internet, listened to streaming audio, ran speed tests from giganews.com, and downloaded the original 30MB file from the other SM, all at the same time. My speed meter (NetStat Live) showed that my incoming and outgoing bandwidth were around 3 Megs, which is about correct based on a 50% Downlink Data Percentage. But, it seems as through you have to really load the APs with traffic to achieve this bandwidth. When you think about it, the SM is getting the 3 Megs in each direction, but that same 30MB file that I was downloading is probably still transferring at the same rate it was, maybe even slower, prior to changing to 50%. Why isnt the full bandwidth allocated to a single file transfer?!? Is this a characteristic of Software Scheduling?
Software scheduling, because of the way it works, ends up limiting an SM to about ~4 Mbps of aggregate (up plus down) throughput. With software scheduling it takes 2 SMs to load the AP up to its ~6.2 Mbps of aggregate throughput.
On the other hand, with hardware scheduling a single SM has access to the full AP aggregate throughput of ~7 Mbps in 1X operation, and ~14 Mbps in 2X operation.
So just to confirm, an Advantage SM connected to an Advantage AP, both running in 2x mode, will allow the SM to realize full 14 Mbps throughput, without having to overload the SM or AP with maximum traffic? I would be able to download a file from one machine connected to an Advantage SM from another machine connected to an Advantage SM at 14 Mbps?
Or would this sort of be like a 20 Mbps backhaul, which has 14 Mbps throughput, and the actual speed would depend on the Downlink ratio?
So just to confirm, an Advantage SM connected to an Advantage AP, both running in 2x mode, will allow the SM to realize full 14 Mbps throughput, without having to overload the SM or AP with maximum traffic?
Yes, 14 Mbps aggregate (up + down) througput.
I would be able to download a file from one machine connected to an Advantage SM from another machine connected to an Advantage SM at 14 Mbps?
No. 7 Mbps would be the max. You have to split the 14 Mbps between up and down. See below for more info.
Or would this sort of be like a 20 Mbps backhaul, which has 14 Mbps throughput, and the actual speed would depend on the Downlink ratio?
It is very much like the 20 Mbps backhaul. You set the up/down split of the 14 Mbps aggregate by setting the Downlink Data % for the sector.
When using a Canopy network for Internet access, 75% down (25% up) works well, as users typically download a lot more than they upload.
When using a Canopy network as a true network, with SM to SM traffic like you describe predominating, 50% may be a better setting. In your case, if you set an Advantage AP to 50% Downlink, a file will move between two SMs at around 7 Mpbs, assuming they are series P9 hardware, have good enough RF to operate at 2X, and have the AP to themselves at least at the instant in question.
Note, if these are standard SMs, the burst rate will be 7 Mbps but after you exhaust the token bucket (set by the burst allocation setting), the rate will drop to 3.5 Mbps (or whatever you have set the sustained rates to). If they are Advantage SMs, they can run full out forever at 7 Mbps.