I did some repeat testing, and that 640mbps was a bit of an outlier. Repeat tests on fast.com were closer to 450, and my CPU was pegging close to 100%.
So if you are getting 450 also on 8.2.1 that sounds about the same as me…
Just tested speedtest.net and got 480, then 476:
I tested on 8.4.0 RC 1. The results were the same. I then posted the following on Firmware 8.4.0 RC 1:
There is definitely a real world performance problem with Balance One Core with 8.4.0 RC 1 (build 5585) not reaching is rated 600Mbps stateful firewall specification. It is reaching 300Mbps on Ookla’s Speedtest.com, the gold standard speed test for most people, and 450Mbps on Fast.com which increases the size of their average payload. Ookla sticks with the MTU size, a reasonable real world test approach.
This has been verified for multiple Balance One Core routers, and by more than one person. I bench tested three different Balance One Core routers, after resetting to factory defaults, on a 1.2Gbps Comcast line using a Firewalla Gold Plus in between a 1Gbps modem and the Balance One, then in between Surf Soho. The Firewalla easily reaches 900Mbps. The Surf Soho reaches 120Mbps on Ookla and Fast, its rated speed. The Balance One Core is reaching 50% of its rated speed on Ookla and 75% of its rated speed on Fast.com, the same results as when directly connected to the 1Gbps modem.
The difference between the Surf Soho and the Balance One Core is the CPU usage. This is particularly evident when running the Firewalla internal speed test where the Surf Soho can reach 140-150Mbps at 70-80% CPU while the Balance One Core can reach 550Mbps to 650Mbps at 90-100% CPU. The Firewalla uses much bigger packet sizes than the modem MTU to achieve this.
I also discovered that the difference in speed tests results is highly dependent on what the router is doing internally. I.e., at idle the Peplink routers CPU can range from 17% to 35%. And I always thought it was Comcast that was causing variable speed test results!
I hope that Peplink engineers can rapidly confirm the Balance One Core speed test results, and then let us know what the next steps will be. Peplink has always been conservative in their numbers, so my hope is that there is some bug that is easily fixed, although my testing on 8.3.0 and 8.2.1 produced similar results, so this is not unique to 8.4.0 RC 1. Or will this take longer to remedy? Or worse, will we have to live with the reduced Balance One performance on version 9 and beyond?
Just a quick dotting-the-i’s question: Is your Balance One by any chance licensed with the 5-WAN upgrade (BPL-ONE-LC-5WAN)?
That drops the throughput to 400Mbps.
Cheers,
Z
They are not. Good fact to know though! Thanks
I would not put too much faith in the CPU meter. My Core 1 pegs the CPU meter at 100% at 350 Mbps, so ignore it.
My Core 1 device has streamed at 803 Mbps at Ookla (single socket) last week, and regular tests at 600+
There is something dodgy going on. We have a Balance20X which appears to be the weak link in an office brance. Even if you look past the unfortunate marketing “1Gbit stateful firewall (as long as you don’t use any features)” we can’t get it to push higher thatn 200kB/s upload on WAN.
This is an image of me trying to download some of our footage from our remote server bahind the balance, plugged in directly to LAN1:
We have a range of other objects behind the balance we were hoping to have since upgrading to the Balance 20x:
None of these are working.
Our ISP is a business line providing 400/40 uncapped without traffic shaping. Speed tests on site show less than perfect results on copper (100/30) and are much better on WIFI (150/35) when connected to the Balance internal AP.
The only features we are using on the Balance at the moment is the firewall and the AP. No QoS or anything like that.
I should add, that when downloading that same file over LAN it downloads full speed.
I’d say it’s screwed.
Are your WIFI clients on the same network?
We’re seeing the odd things in our office with this model too. Speeds don’t make sense.
We found LAN1 to LAN2 when using VLANs cripples speeds between the two ports.
I don’t know if this is the firmware or infact poor hardware. I’ve still got a ticket open with Peplink after one year trying to return our Balance20X as they were suffering memory leaks and not getting close to the advertised recommended users.
We were loathed to believe our issues were with the Balance 20X having upgraded to it so to make it a fair experiment, the tests were conducted in isolation - no interference from other clients/networks.
So I tried what you said and tested that just now and it’s exactly what you said.
Over WAN to the server I got 85kB/s
Over VPN to the server I got 220kB/s
Then I VPN’d to a test machine on the same network as the server and got 40MB/s
I put the test machine on a new VLAN/network and got a very unimpressive 4MB/s
The test machine and the server are on the Balance lan ports 1&2.
Where to go from here?
The 100/30 results you report are suspicious because 100Mbps quite often means that there is a bad Ethernet cable or other issue where the connection is at 100Mbps rather than 1Gbps. You can look at the port speeds under Status->Ethernet Ports, then click on each of the ports to see what speed they have negotiated. Your server will also have the ability to display its port speed.
If you have a smart switch laying around that passes VLAN tags, you can hook up your test machine and server to it to see what speeds you get. Note: I believe that the Balance 20X doesn’t have an internal switch, but rather runs LAN to LAN traffic through its CPU. So a smart switch can significantly offload the 20X if there is a lot of LAN to LAN traffic.
Edit: If you have a dumb switch, you can change the 20X port it is connected to from Trunk mode to Access mode with the appropriate VLAN ID.
What firmware version are you on where your “Core 1 device has streamed at 803 Mbps at Ookla (single socket) last week, and regular tests at 600+”?
FYI, I just upgraded to 8.4.0 RC 2 and retested. No difference from prior testing. And no significant difference on Ookla speedtest.net between Multi and Single streams (but it is the middle of the day when things are busier on Comcast).
8.3.0 build 5514. https://www.speedtest.net/result/14375998945.png
Account is supposed to be pegged at 400, but some days the limits come off. I have about 70 Outbound policy rules, and run the PepVPN to a Fusion hub elsewhere, for minor things. Not much else.
Since we’re having fun - let’s remove the middleman:
I don’t have a Balance One lying around. Ours are deployed on slower lines elsewhere, but I do have a B20X test unit, and since that was a model that came up in the conversation I might as well try a direct unit-to-unit speed test.
Test setup:
Test:
Run the WAN analysis on the B20X (keeping in mind that the BR1 is busy with a slew of other clients besides the B20X):
Upload from the B20X:
Download to the B20X:
On another occasion I have run the same test between two routers (a B20X and a B380 HW6) that were not busy with anything else, and the link was then steady at close to 1Gbps).
If possible, run a test like this, without other devices in the middle creating noise, and not being dependent on the peculiarities of the various speed-test sites (e.g. fast.com seems particularly confused about multi-WAN routers).
Cheers,
Z
WAN Analysis is excellent for establishing WAN capabilities. It does not include any LAN traffic though, so I discovered that it isn’t an appropriate test for Balance One throughput as it takes CPU to facilitate WAN to/from LAN transfers. While I hope WAN Analysis was doing Stateful Firewall checks, I can’t verify that. Thank you for running B20X tests!
On WAN Analysis between two Balance One’s (connected by 1Gbs switch), I get around 590-600Mbps. If I have a second web page up with the Dashboard to display CPU usage, this drops 10% to 540-550Mbps, once again pointing to the CPU usage on the Balance One Core being a bottleneck. This is on Balance One’s which have been factory reset running 8.3.0.
My conclusions to date are that the Balance One cannot consistently deliver more than 450Mbps on the fast.com speedtest and 300Mbps on the Ookla speedtest (everyone’s standard speedtest). It is possible to get an outlier event occasionally when the CPU isn’t running some sort of internal task. The best I’ve ever seen in weeks of testing is 530Mbps on fast.com (and that was without being logged into the Balance One).
The Surf Soho can consistently reach it’s rated 120Mbps. If you have a moment, I would appreciate if you will you please test the B20X with Ookla’s speedtest.net and fast.com.
Establishing a proper test environment can be tricky.
With the multi-hop tests (such as Speedtest and fast.com) you are subject to whatever is between the router and the test server. With a Peplink-to-Peplink connection you at least are in control of your test environment. E.g. you can load either end of the connection as you wish, such as having active clients on the WAN or LAN at either end at the same time you perform the test, and similarly you can load up either end with outbound policies and whatever else you want to test against. (E.g., the server used in my example was busy with other clients, and may have underperformed because of that).
The key thing is that you need to control all aspects of the test harness, particularly if you are claiming that the vendor is misrepresenting the specs.
FWIW., in a clean (unloadded) environment I have seen Peplink devices outperform the claimed specs w.r.t. throughput. As one loads other tasks on the router then it may perform at a lower level. The specs do not address that. Empirical data identifying contributors to a lower bandwidth (or other functionality) is helpful, but does not per se invalidate the specs as provided.
Cheers,
Z
The interesting portion of my testing is with a “Balance One connected to a LAN port of a Firewalla which is getting 900Mbps+ from the modem but that didn’t help either” (done as a double check against my vanilla modem to Balance One testing). I.e., the Firewalla is getting fed at 900Mbps+ from Ookla speedtest.net and fast.com, and then feeding the Balance One as fast as the Balance One can ingest. Of course, there is a limit to the buffer size in the Firewalla and ACK’s from the Balance One can have some effect on the transfer rate after a certain number of ACK’s are outstanding. However, you would expect the Balance One to perform at least somewhat better if multi-hops were affecting results, but it didn’t. (Again, this testing is with nothing else going on and with the Balance One factory reset so no other tasks.)
I too have found Peplink to be conservative in their specifications which is why I checked a Surf Soho and it met spec. While I didn’t check an older B20 which I have gathering dust, I suspect it too would meet spec. My expectations for the B20X is that it too will meet spec, hence my request for B20X testing with Ookla speedtest.com and fast.com from you or others. If the B20X meets spec, and my testing as well as others can’t get the Balance One to consistently meet spec, it would seem to cast doubt on whether the Balance One can meet spec.
I have another Balance One in production which should be replaced if it is not capable of meeting spec, perhaps with a B20X (especially if the B20X meets spec). That is in addition to one of the three Balance One Cores which I have been testing being destined for an installation which requires 600Mbps. My double checking to make sure it could support 600Mbps before I installed it is what lead to this post. I may have to purchase a B20X instead just to get 600Mbps if I and others can’t consistently get 600Mbps from the Balance One which is why I am spending so much time and effort on this.
Hi Mark,
I suspect there’s less to it than that.
The test machine was plugged into LAN port 2 on the B20X.
The server is plugged into LAN port 1 on the B20X.
The test machine downloaded that file from the server using a browser. When the test machine is on the same VLAN, it managed 33.7MB (image1). But when the test machine was added to a new network/Access VLAN you can see the speed is really bad @ 4.1MB (image 2) Cables are factory made and proven with the Fluke to 10Gbit.
The WAN connection negotiated at 1Gbit, as do the LAN ports and can reach over 100Mbit. But this test I did I think is showing something wrong with the device ability.
Laptop ↔ B20X ↔ Server
Ok, as a starting point I understand that you have verified ethernet negotiated at 1Gbps for the WAN and LAN ports. Thus you can run a speed test like Ookla speedtest.net or fast.com from both your server and your test server over the WAN and get close to 400/40.
I would next check the routing of the packets. I believe you expect the test machine to download the file from the server without going over the WAN, simply transferring through the LAN ports. Can you verify that this is happening? One way would be to simply unplug the WAN when running the test. However, I believe you aren’t onsite and are remotely conducting the tests. So what happens to the WAN usage as shown by the B20X (Status->Usage Reports Real-Time)? You might also try tracert to see the hops which are occurring.
It also never hurts to run a file copy directly between the two machines to see what happens.
There is the Balance 580 and 580X, but those are a tad more expensive than the Balance One.
Or you can work with synergy mode, adding multiple routers into one composite. E.g., we have one installation where a Balance 20X (one 1Gbps WAN) is coupled with a transit duo in synergy mode (+2 WANs, in this case cellular. Substitute an HD2 for the transit and you get two additional wired WANs) and an HD2 (not new…) on a virtual WAN (+2 WANs), ending up with a 1 Gbps router (the B20X) with three WANs managed locally and additional 2 managed separately. It is klugy, but with spare equipment lying around it is quite workable 
Having written that, since your one WAN is at 940Gbps it really does not make too much sense to pile on the others, unless there is a redundancy factor involved. Five different modes of connection?
Cheers,
Z
This is turning into a bad case of subject creep. Please open a new topic with a new subject title, and I’ll respond there (or send me a PM for a personal communication).
Cheers,
Z
Finally achieved 600Mbps download on Ookla speedtest.net through Balance One; 767/174 Mbps was the high after multiple tests on multiple servers. The key was using the Edge browser (Win 10) instead of Chrome, configuring speedtest.net to Single Connection instead of Multi Connection, and I did not log into the Balance One (the Dashboard display uses CPU). Selecting Single Connection severely limits the upload speeds for some unknown reason. Upload speeds increased to a high of 567Mbps on one of the Multi Connection tests.
Thanks to @soylentgreen, @rossh_pl_beta and @zegor_mjol for your responses which helped guide me to this resolution.
