Speedfusion degrades performance

Paille · January 7, 2015, 9:05pm

Ok, it works with this one

thanks

Platts · January 27, 2015, 2:03am

How can we access the WAN Smoothing functionality?

TK_Liew · January 27, 2015, 4:30pm

Please find the attached screen shot. It called “Packet Loss Compensation” in v6.2b03.

countryman · June 16, 2016, 8:27am

Has this problem been solved?
I am trying to find a solution for the same problem for more than a month with Support (ticket #762287).

Have changed all the hardware (new modems, another Peplink server), have changes ISP - all useless: combined speed decreases dramatically from (20+15+20+6 mbps) to 2-4mbps, and nothing could be done!!!
Have tried several options (wan-smoothing, latency cut-off, different WAN priority, MTU change, different IP-range, different DNS etc).

Tech support is silent more than a week. I purchased this device especcially for broadband bonding, but now i use only for simple balancing (have no speed increase, have no failover-unbreakable option and my IP-phones and live video-streaming sometimes fail).

TK_Liew · June 16, 2016, 1:35pm

Sorry to keep you waiting. We have notified the support engineer on this. Your case will be escalate to engineering team for further checking. They will get back to you soon.

CA_Tallguy · October 22, 2016, 4:07am

Sorry to bump an old thread but while I am impressed with the product I too wish that the units could do a better job at recognizing when a connection is dragging down performance of others and conversely which WAN’s should be left in the mix. Basically I want the software to figure out where the problems are rather than me having to do this manually. Perhaps this couldn’t happen in milliseconds as some trials and history might be needed. Perhaps some kind of “scoring” system where each WAN is continually graded with weight given to the newest grades. Points added for throughput, low latency, etc and taken away for lost packets and low throughput in relation to other WAN’s. Something like that. Software might do tests on an interval, trying different WAN combinations to see if performance increases with one or more WAN’s put into hot failover or standby. If performance is good then give WAN’s involved points and take away if bad. Or just isolate the problem WAN’s by trial and error in some way.

As others have said, we really need the software to automate what we have to do manually right now. Some of our WAN’s may go in and out of 2G 3G 4G etc so the suggestion to use 3G’s as hot failover and LTE’s as primary won’t always work. Where I am now, one provider only has 3G but the next location will have LTE and another provider might even flip to 2G. So I find myself manually enabling and disabling the WAN’s when I would love to be able to just put in 5 WAN’s and let the software figure out what is best. Failure checks some of the work and it seems like SpeedFusion sometimes will stop sending traffic on a WAN when performance degrades (or else no traffic is flowing because of the connection, not sure), but as others have said – it needs to figure out if there is a “problem” WAN or or more dragging down what otherwise would be exceptional performance – and then put that WAN on standby.

CA_Tallguy · October 22, 2016, 4:15am

Basically if SpeedFusion were the coach of a basketball team – it needs to figure out which players should be in the game and which should be on the bench. Right now it mostly just knows if a player is dead LOL.

Steve · October 24, 2016, 8:17pm

CA_Tallguy:

Sorry to bump an old thread but while I am impressed with the product I too wish that the units could do a better job at recognizing when a connection is dragging down performance of others and conversely which WAN’s should be left in the mix. Basically I want the software to figure out where the problems are rather than me having to do this manually. Perhaps this couldn’t happen in milliseconds as some trials and history might be needed. Perhaps some kind of “scoring” system where each WAN is continually graded with weight given to the newest grades. Points added for throughput, low latency, etc and taken away for lost packets and low throughput in relation to other WAN’s. Something like that. Software might do tests on an interval, trying different WAN combinations to see if performance increases with one or more WAN’s put into hot failover or standby. If performance is good then give WAN’s involved points and take away if bad. Or just isolate the problem WAN’s by trial and error in some way.

As others have said, we really need the software to automate what we have to do manually right now. Some of our WAN’s may go in and out of 2G 3G 4G etc so the suggestion to use 3G’s as hot failover and LTE’s as primary won’t always work. Where I am now, one provider only has 3G but the next location will have LTE and another provider might even flip to 2G. So I find myself manually enabling and disabling the WAN’s when I would love to be able to just put in 5 WAN’s and let the software figure out what is best. Failure checks some of the work and it seems like SpeedFusion sometimes will stop sending traffic on a WAN when performance degrades (or else no traffic is flowing because of the connection, not sure), but as others have said – it needs to figure out if there is a “problem” WAN or or more dragging down what otherwise would be exceptional performance – and then put that WAN on standby.

Thank you, we appreciate comments from users like you so we can know more about the real problems and improve our features.

Sending test traffic has been an idea raised several times within the SpeedFusion team, however there is no strong evidence yet this has more advantage compared to the current mechanism - using user’s traffic to determine the quality of the connection. The SpeedFusion core engine always monitor the latency and packet loss rate of each connections, and will temporarily suspend those that are pulling down the performance, and actually we have exposed some of the parameters on Web Admin so users can fine tune the value, many of them is available since firmware 6.2.0, but we are still revising them and giving you more control. For example, in firmware 6.3.2, we have added “Latency Difference Cutoff” (it’s an hidden option and can be enabled by clicking the help icon on PepVPN profile panel, help text: “Traffic will be stopped for links that exceed the specified millisecond value with respect to the lowest latency link. (e.g. Lowest latency is 100ms, a value of 500ms means links with latency 600ms or more will not be used)”) so users can control how connections are temporarily suspended when the latency becoming worse.

As for the 4G/3G selection, though we don’t have an automated way for this yet (but yes, it’s on our road map), one suggestion could be making use of latency, since 3G usually has higher latency than 4G, may be setting Latency Cutoff value for this WAN will help to improve the performance. This is not perfect, we do understand and will keep improving SpeedFusion in future firmware releases.

CA_Tallguy · October 25, 2016, 2:25am

Thank you for the detailed response! I appreciate all the development in the product and it is 100 times better than my previous solution from provider “CP”

I like that SpeedFusion does try to figure out best configuration on the fly but it does not seem to figure out which links are dragging down in my situation. Perhaps it is because I am a single user and my traffic is thus very sporadic rather than steady? The other thing I wonder is if it is not figuring out which links have more capacity that could be used or that capacity could be used differently. The result to me usually feels like my NET performance is the average of all the links. There doesn’t seem to be too much additive.

I end up having to enable and disable links trying various combinations to see what offers the best throughput. I have 5 links currently: 2 from T-mobile via USB, 1 from T-Mobile via wifi-wan, 1 sprint (3G at this location) and 1 ATT via USB. All are LTE other than Sprint. I just spent 15 minutes trying various combinations and I think leaving in just the 3 LTE links via USB provide the best performance. 12Mbps+ vs 2Mbps or so with all 5. The Sprint 3G connection is pretty decent for maybe .5 to 1 Mbps but that’s it. Sometimes it gets 2Mbps. Here is a summary of the individual links:

USB1: T-Mobile, 10x3Mbps, speed goes up during test, 75ms latency
USB2: T-Mobile, 6x3Mbps, speed ramps up during test, 78ms latency
USB3: AT&T, 7x11Mbps, burst for first 2 or 3 seconds to 10 or 11mbps, 107ms latency, small amount of packet loss at saturation
USB4: Sprint 3G, .75x.75, heavy packet loss at max but just low when 200-500k is being used, 171ms latency
WIFI: T-Mobile, 5x1.5, 114ms latency, no packet loss but less consistent link since it is wifi at some distance

Performance is best with just links 1&2 and maybe also 3 but still those 3 together just gets me to about 10x3mbps. That is with WAN smoothing turned off.

If I toss all the links in the mix I get about 3x1 connection effectively. Like I said above, it seems like the average of all the links (maybe even less I guess since average would be about 6mbps down) with the 3G link really pulling down the average. The quality of the bonded link is pretty good, however. What I think it should be doing is setting USB4 and WIFIWAN aside for hot failover (not marking them as down)… or just using the capacity they have, maybe just for an error correction channel? (I don’t know if lagging error correction data is useful, however.) I would think that it should be able to remove the 3G connection simply because it has not observed any throughput over 1mbps and the others are far faster, but why is it not figuring this out?

On the other hand, I like that my poor old 3G connection under most usage is pulling the same amount of weight as the other links. I watch the speed fusion status as I surf and I get a good couple 100 K out of the link doing normal activities. So I don’t feel like I am wasting money for that bandwidth with the connection sitting unused. In the end, it isn’t so bad that throughput is 3x1 or whatever because it helps keep overall bandwidth usage in check. 3x1 is plenty of performance for most of what I do. But it does seem to not make much sense when 4 of the 5 links have individual performance that is better than that.

TK_Liew · October 25, 2016, 12:14pm

CA_Tallguy:

Thank you for the detailed response! I appreciate all the development in the product and it is 100 times better than my previous solution from provider “CP”

I like that SpeedFusion does try to figure out best configuration on the fly but it does not seem to figure out which links are dragging down in my situation. Perhaps it is because I am a single user and my traffic is thus very sporadic rather than steady? The other thing I wonder is if it is not figuring out which links have more capacity that could be used or that capacity could be used differently. The result to me usually feels like my NET performance is the average of all the links. There doesn’t seem to be too much additive.

I end up having to enable and disable links trying various combinations to see what offers the best throughput. I have 5 links currently: 2 from T-mobile via USB, 1 from T-Mobile via wifi-wan, 1 sprint (3G at this location) and 1 ATT via USB. All are LTE other than Sprint. I just spent 15 minutes trying various combinations and I think leaving in just the 3 LTE links via USB provide the best performance. 12Mbps+ vs 2Mbps or so with all 5. The Sprint 3G connection is pretty decent for maybe .5 to 1 Mbps but that’s it. Sometimes it gets 2Mbps. Here is a summary of the individual links:

USB1: T-Mobile, 10x3Mbps, speed goes up during test, 75ms latency
USB2: T-Mobile, 6x3Mbps, speed ramps up during test, 78ms latency
USB3: AT&T, 7x11Mbps, burst for first 2 or 3 seconds to 10 or 11mbps, 107ms latency, small amount of packet loss at saturation
USB4: Sprint 3G, .75x.75, heavy packet loss at max but just low when 200-500k is being used, 171ms latency
WIFI: T-Mobile, 5x1.5, 114ms latency, no packet loss but less consistent link since it is wifi at some distance

Performance is best with just links 1&2 and maybe also 3 but still those 3 together just gets me to about 10x3mbps. That is with WAN smoothing turned off.

If I toss all the links in the mix I get about 3x1 connection effectively. Like I said above, it seems like the average of all the links (maybe even less I guess since average would be about 6mbps down) with the 3G link really pulling down the average. The quality of the bonded link is pretty good, however. What I think it should be doing is setting USB4 and WIFIWAN aside for hot failover (not marking them as down)… or just using the capacity they have, maybe just for an error correction channel? (I don’t know if lagging error correction data is useful, however.) I would think that it should be able to remove the 3G connection simply because it has not observed any throughput over 1mbps and the others are far faster, but why is it not figuring this out?

On the other hand, I like that my poor old 3G connection under most usage is pulling the same amount of weight as the other links. I watch the speed fusion status as I surf and I get a good couple 100 K out of the link doing normal activities. So I don’t feel like I am wasting money for that bandwidth with the connection sitting unused. In the end, it isn’t so bad that throughput is 3x1 or whatever because it helps keep overall bandwidth usage in check. 3x1 is plenty of performance for most of what I do. But it does seem to not make much sense when 4 of the 5 links have individual performance that is better than that.

I suggest open ticket for us to check the bahavior of each cellular connection . Then we can provide the better advise.

cellworld · January 25, 2017, 2:13am

100 times better than CP, thought this was a PepLink forum ha ha always nice to see what looks like a PepLink employee adding Jerrys Final Thought

CA_Tallguy · April 1, 2017, 8:10am

Just a follow up – my situation has improved dramatically with recent firmware updates and after much assistance from the engineering team. Also, what made a BIG difference is enabling WAN smoothing on the FusionHub side. Now I’m in a situation where I need to throttle users so we don’t blow though our bandwidth too quickly! Good problem to have.