B380
• Firmware 7.1.0
• One NAT WAN port to internet is 10MB up and 40 down
• One Lan port has gigabit switch with one Windows 2008R2 64 bit server running ftp server. Nothing else on LAN.
• DHCP on for LAN
• Pepfusion is on but AES-256 is off a
• Wan Smoothing is off
• MTU 1428
HD4 – LTE modems
• Firmware 7.1.0
• One LAN port has gigabit switch with one Windows 7 PC. WINscp Version 32 bit 5.13 installed with logging turned on. Nothing else on LAN.
• DHCP on for LAN
• Data SIMS:
o 4 SIMS in the slot A position, none is slot B
o All 4 in bond
o 2 Vodafone, 1 EE, 1 O2.
o At time of test all were LTE in a strong signal area.
o PepVPN test showed around 22 MB upload to B380
o All set for priority 1 connection.
o All had MTU 1428
o NAT on
• No Other WAN connections
• Pepfusion is on but AES-256 is off
So we had an image file that was 10,561,547 bytes and did not compress well ( standard zip gave 10,550,475 bytes). We made 5 copies of this file under different names and the opened a WINScp connection back to the FTP server and dragged/dropped these 5 copies to ftp server in one select drag/drop. Took 1 minute 4 seconds.
Then we opened 4 more WINscp to same FTP server on same PC and created 5 folders (1 to 5) on the FTP server. We then did a drag/drop from the PC to the FTP server in each window copy 1 to folder 1 and so on. Of course, we had to click between windows to initiate the drag drop but this only about 6 seconds. All 5 completed in 23 seconds from first drag/drop initiated.
In the first test the five files were sequential and in the second test they were roughly in parallel.
We tried it again 10 minutes later and got roughly the same ratios.
My issue is why was the second test was so much quicker? I understood the whole point of SpeedFusion was to spread the load across the bond.
Some questions.
What is the average latency between the B380 and the HD4?
What FTP server software is installed on the Windows 2008r2 box (is it the MS IIS FTP role service or a dedicated 3rd party FTP server application)?
What speed / times can you achieve for a single file FTP upload to that Windows server from a local device on the same LAN segment?
When doing a single file copy over FTP did you watch the SpeedFusion Status pages on the two devices? I assume you saw traffic being used across all WANs in that view?
The diagram here is a common one to compare the VPN’s. The last two rows confuse me a little.
“SpeedFusion WAN Smoothing” does not use “Bandwidth Aggregation”. It says that “SpeedFusion WAN Smoothing” uses VPN Bonding Across Multiple WAN Links. So with Wan smoothing on what’s the difference? I have come across the statement below re WAN smoothing where the part in bold implies use of algorithm similar to “Lowest Latency”.
"Using intelligent algorithms, the sending Peplink device builds and delivers special packets. Armed with these special packets, the receiving Peplink device can then reconstruct the lost packets to ensure that communications remains consistent. At the same time, WAN Smoothing will attempt to assign traffic to the WAN connection with the lowest latency. Thus, the latency of the SpeedFusion tunnel becomes the latency of the most responsive WAN connection.
Similar to Bandwidth Bonding, WAN Smoothing combines the bandwidth of the available WAN connections. The major difference is that WAN Smoothing prioritizes connection consistency (reduced packet loss) rather than on increasing connection speed"
Good stuff, looks like you’re isolating the issue.
Is it passive mode or server mode FTP?
In server mode filezilla advertises a 4Mb TCP Receive window which I know can max a 100Mb fiber link at 300ms.
In passive mode the client initiates the FTP TCP session and so sets the TCP receive window size which might be considerably lower than 4Mbs (ie 64Kb).
WAN smoothing duplicates traffic across multiple WAN links to mitigate WAN links with jitter and packet loss issues. This process consumes bandwidth so the link will always have less bandwidth that the total available bandwidth available across all the links.
WAN bonding is bandwidth aggregation at a packet level across available WAN links.
Perhaps an example will bring clarity. Picture a scenario where real time services (like a VoIP call) are in use on a MAX HD2 with two cellular WANs. If you were using SpeedFusion Bonding across these links (where one stream is load balanced across the WANs at a packet level within the VPN tunnel), the latency on that VPN connection will vary as the cellular latency varies across both links. This would mean that latency would rise on the tunnel as a whole even when the latency on a single cellular WAN rises.
With WAN smoothing, the stream is replicated across both WANs, and the packets that arrive first at the remote end are used to rebuild the stream. So if only one cellular links latency rose, the overall tunnel latency would remain the same as the lowest latency WAN link.
The compromise with WAN smoothing of course is that you are doubling the amount of bandwidth used to send a single stream of data.
