On Mon, Jan 30, 2017 at 6:38 PM, Nathan Anderson <[email protected]> wrote:
> ...though I should warn you that we are using epc-access-if-list as > detailed in past posts of mine, so we are unable to have our ENBs talk to > our EPC over the management VLAN. Thus we can only run iperfs between ENB > and EPC over the S1 VLAN, which isn't something you can normally do (due to > iptables filter rule installed by Telrad on EPC side). > I had intended to use epc-access-if-list, but I wasn't able to make that work and still have L2 redundancy. I might have missed something obvious? Honestly I'd be fine with a LAG, but I wasn't 100% sure it is supported yet. (Or better yet, a 10G interface.) > I am pretty sure that that filter rule wasn't there previously (probably > in initial 6.5 release), and although I cannot recount the details, I also > seem to recall that we saw significant differences between iperf tests over > the management VLAN and iperf tests over the S1 VLAN. I want to say the > tests over the management VLAN were notably worse, and since we knew > traffic between devices was not running over that VLAN anyway, we didn't > care and discounted those results. At the time I figured that Telrad had > optimized for performance over the S1 VLAN somehow at the expense of the > management one; nothing else could make sense of the difference. > Because we're using L3 EPC<>ENB and the VLANs are only relevant at the last hop, I was assuming that performance would be the same on mgmt vs. S3. And Telrad did iperf testing on my network *using management*, and those were the results I was trying to replicate. So if I understand you correctly, you haven't tested iperf to management on both L2 and L3? This would have to be contrasted with S1 (again, L2 and L3), which you can't do with your setup. > > We also switch (well, okay, VPLS) all traffic between ENBs and the EPC so > (from the ENB and EPC's perspective) there is no L3/routed infrastructure > sitting between them. So the way we have our network set up may not be an > apples-to-apples comparison with yours. > > > I'm leaning toward this topology, unless S1 routed performance is identical to switched — but this isn't what Telrad tested! What Telrad was trying to prove was that S1/backhaul performance is inadequate and causing some performance issues. I was unconvinced, as the results were significantly poorer than router-to-router testing. And in attempting to replicate them, I discovered that iperf using management is radically different on L3 vs L2. > -- Nathan > > > > *From:* [email protected] [mailto:[email protected]] *On > Behalf Of *Nathan Anderson > *Sent:* Monday, January 30, 2017 7:21 PM > > *To:* [email protected] > *Subject:* Re: [Telrad] Issues > > > > Oh I see, gotcha. Happy to help off-list if that's where you want to take > it. > > > > -- Nathan > > > > *From:* [email protected] [mailto:[email protected] > <[email protected]>] *On Behalf Of *Jeremy Austin > *Sent:* Monday, January 30, 2017 7:19 PM > *To:* [email protected] > *Subject:* Re: [Telrad] Issues > > > > Nathan, I can send you more details off list. I should have been clearer; > I'm getting different results when testing iperf on L2 to the EPC > management network vs. L3 to management, even when taking bandwidth delay > product into account. > Thanks... more later. > > On Mon, Jan 30, 2017 at 6:08 PM Nathan Anderson <[email protected]> wrote: > > Can you detail your specific results? > > > > I'll try to run some iperfs between our ENBs and EPC so we can compare to > yours, but we are not doing L2 PDNs so not sure how much help we can be > there. > > > > -- Nathan > > > > *From:* [email protected] [mailto:[email protected]] *On > Behalf Of *Jeremy Austin > *Sent:* Monday, January 30, 2017 7:01 PM > > > *To:* [email protected] > *Subject:* Re: [Telrad] Issues > > > > > > On Sun, Nov 20, 2016 at 2:13 AM, Nathan Anderson <[email protected]> wrote: > > Which again leaves us at square one as far as explaining what is going on, > but at least we now have a working solution. So, huzzah for that. Ever > since moving the EPC trunk port to a different port on our LTE core > router, I have been able to do 60Mbit/s single-stream TCP tests to UEs that > are multiple hops out on our microwave backbone. I have NEVER been able to > accomplish this before, much less consistently. I think that using the > word "giddy" here to explain our disposition ever since we made this > discovery would be a vast understatement. > > > To revisit an older thread, has anyone else done iperf tests between ENB > and EPC and found a bottleneck that wasn't a backhaul? I'm getting > radically different L2 vs. L3 results, and it does seem to be narrowed down > to the EPC. > > > > S1 and PDNs are on the same switch port. > > > > > -- > > Jeremy Austin > > > > (907) 895-2311 > > (907) 803-5422 > > [email protected] > > > > Heritage NetWorks > > Whitestone Power & Communications > > Vertical Broadband, LLC > > > > Schedule a meeting: http://doodle.com/jermudgeon > > _______________________________________________ > Telrad mailing list > [email protected] > http://lists.wispa.org/mailman/listinfo/telrad > > > _______________________________________________ > Telrad mailing list > [email protected] > http://lists.wispa.org/mailman/listinfo/telrad > > -- Jeremy Austin (907) 895-2311 (907) 803-5422 [email protected] Heritage NetWorks Whitestone Power & Communications Vertical Broadband, LLC Schedule a meeting: http://doodle.com/jermudgeon
_______________________________________________ Telrad mailing list [email protected] http://lists.wispa.org/mailman/listinfo/telrad
