XG-7100 efficiency low?

johnpoz

Before I did test I made sure that my public facing IP is different on each connection.

And you validated that your routing is working as you assume.. Via checking the state tables and traffic flow over both wan side interfaces..

Something is off that is for sure - the thing is way more capable of just 1 gig ;)

Now that @stephenw10 sure can get to the bottom of the issue.. I don't have a XG7100 to play with :( or would be happy to duplicate your testing...

mke

I did not checked state tables, only did "what is my public IP" check in the browser.
I have opened ticket with support, sent them status dump for my unit. Waiting for the issue to be potentially replicated, let's see they said it may be days.

johnpoz

Support never shares anything about any specific issues that get moved to support tickets, even when the thread was started in the forum..

So hope you will share what the issue is, when figure it out. Best I could do would be to duplicate the setup you are using for testing with my sg4860.. But that wouldn't be much help, since doesn't have any switch ports.

mke

Will share info whenever they come with some answer.

johnpoz

Great.. Got to be something stupid ;) I don't think it will be "days" either..

Rico

Is your problem fixed?

-Rico

johnpoz

I had heard a snip from Chris that they were about to discuss with the ticket opener.. But that is all I got out of him ;) Hope the OP comes back and let us know some info about this myself.

MalnPr0

I setup a lab to test this. The LAGG appears to be working very well.

Lab Details:
HOST-1 (SG-5100)
DUT (XG-7100)
HOST-2 (SG-5100)

HOST-1
IX0 -> LAGG0.4091 / PORT 1 (ETH1)
IX1 -> LAGG0.4090 / PORT 2 (ETH2)
IX2 -> LAGG0.4091 / PORT 3 (ETH3)
IX3 -> LAGG0.4090 / PORT 4 (ETH4)

HOST-2:
IX0 -> LAGG0.3091 / PORT 5 (ETH5)
IX1 -> LAGG0.3090 / PORT 6 (ETH6)
IX2 -> LAGG0.3091 / PORT 7 (ETH7)
IX3 -> LAGG0.3090 / PORT 8 (ETH8)

DUT-NETWORK
LAGG0.4090 = WAN_1
LAGG0.4091 = LAN_1
LAGG0.3090 = WAN_2
LAGG0.3091 = LAN_2

Each WAN and LAN has two IPs assigned from two different networks - for a total of 4 WAN IPs and 4 LAN IPs.

LAN-1 = 1.1.1.1/30, 3.3.3.1/30

16.0.0.0/18 -> 1.1.1.2 (HOST-1.IX0)
16.0.64.0/18 -> 3.3.3.2 (HOST-1.IX2)

LAN-2 = 5.5.5.1/30, 7.7.7.1/30

16.0.128.0/18 -> 5.5.5.2 (HOST-2.IX0)
16.0.192.0/18 -> 7.7.7.2 (HOST-2.IX2)

WAN-1 = 2.2.2.1/30, 4.4.4.1/30

48.0.0.0/18 -> 2.2.2.2 (HOST-1.IX1)
48.0.64.0/18 -> 4.4.4.2 (HOST-1.IX3)

WAN-2 = 6.6.6.1/30, 8.8.8.1/30

48.0.128.0/18 -> 6.6.6.2 (HOST-1.IX2)
48.0.192.0/18 -> 8.8.8.2 (HOST-1.IX4)

For UDP traffic, I used the latest trex build to generate traffic.
For TCP traffic, I used the latest iperf3 build to generate traffic.

Sending 1500 byte UDP packets, I consistently get around 4.8 Gbps (highest was around 4.86 Gbps / 405 Kpps).
Sending TCP packets with iperf over 1500 MTU, I get close to 4 Gbps.

The results of each were the same under the following scenarios (TCP performed a little better with PF disabled):
PF disabled, PF enabled, NAT disabled, NAT enabled, static routes to local WAN, policy routes to external WAN.

In all scenarios, the results line up for both unidirectional and bidirectional (in the case of bidirectional, the same result as unidirectional but the result applies to both RX and TX for each ethernet switched interface).

johnpoz

Yeah those numbers seem what you would think.. So the question now is helping the OP figure out what is going on in his testing.. Prob have to prove to him that something is not wrong with his hardware..

Or what could be in his config that could be causing the problem.

Derelict

I cannot duplicate @mke's findings:

Clients sending (uploading)

Simultaneous iperf3 -c 172.18.208.1 -P4 -t60 and iperf3 -c 172.18.209.1 -P4 -t60

XG-2758 igb1 <-> XG7100 lagg0.4082 <-> lagg0.4083 <-> MacBook Pro 882,881,928 (897Mb/sec)
XG-2758 igb2 <-> XG7100 lagg0.4084 <-> lagg0.4085 <-> Proxmox VM 908,876,895 (893Mb/sec)

Servers sending (downloading)

Simultaneous iperf3 -R -c 172.18.208.1 -P4 -t60 and iperf3 -R -c 172.18.209.1 -P4 -t60

XG-2758 igb1 <-> XG7100 lagg0.4082 <-> lagg0.4083 <-> MacBook Pro 924,924,926 (925Mb/sec)
XG-2758 igb2 <-> XG7100 lagg0.4084 <-> lagg0.4085 <-> Proxmox VM 931,932,926 (930Mb/sec)

MacBook downloading, VM uploading

Simultaneous iperf3 -R -c 172.18.208.1 -P4 -t60 and iperf3 -c 172.18.209.1 -P4 -t60

XG-2758 igb1 <-> XG7100 lagg0.4082 <-> lagg0.4083 <-> MacBook Pro 925,878,899 (901Mb/sec)
XG-2758 igb2 <-> XG7100 lagg0.4084 <-> lagg0.4085 <-> Proxmox VM 860,917,903 (893Mb/sec)

iperf3 servers running on same XG-2758. This is far from a perfect test environment but it is sufficient to duplicate what is being asserted and I was not successful in doing so.

mke

They are still investigating this but so far cannot replicate the problem. Since I have more than one XG-7100 I did more testing but not with iperf but real pipes at two different locations, result was the same(struggling to go over 1gig) and I even did video on this and sent them but can't post since it shows my IPs and I don't have time to do editing.

johnpoz

So can you duplicate their iperf testing?

mke

What I would love to do is to do iperf across real links with multiple sites with mix of xg7100 and SG-8860 all with gig pipes, not sure if will be able and right now I have very limited time.

johnpoz

do you have any of the switch ports still open - you could use those without disruption of your active links.

mke

I don't have access to those devices since they are in different locations so physically it is problematic to do testing right now.

mke

I got finally an answer after multiple test using multiple XG7100, different routers, pipes, core switches, combinations. It ended up that testing in the real world using same website is misleading. I did try different speed tests but wanted to use that same particular on both laptops because for higher speeds it gave me very good results(single laptop) that reflected situation however while testing multiple computers it showed bottleneck on their side, sick.

Derelict

Thank you for coming back and reporting your findings.

That is why one should not rely on external test sites when testing device performance.

johnpoz

@Derelict said in XG-7100 efficiency low?:

That is why one should not rely on external test sites when testing device performance.

You don't say ;) heheheeh