Hi guys.

Anyone has ever heard or achieved a number of 5000 PPPoE simultanous connections on a MikroTik based router, a routerboard or x86 machine??
It would be nice if someone can tell if RB1000 can handle that much of clients. I'm not talking about the database (that maybe will be done with a standalone radius server) but about the simultaneous connections.
I know that the bandwidth used, and other processor hungry tasks on the router are really important, but still an average would be very helpful.


Thanks

5000 pppoe clients on one box? Forget about it. and I mean this seriously.

I by my self think it is not smart - to connect 5000 to 1 device - if device goes down there will be 5000 angry customers. and idea of bridging 5000 clients just scares a crap out of me.

I have no more that 254 (mostly 64) clients per one management router/network. This way I'm able to chop upload traffic as close as possible to the client and this way avoid flooding it into my network.

thx macgaiver for the post.
I was thinking about using 2 or 3 routerboards for that, segmenting the number of users

We use as PPPoE servers FinePoint ServPoET BMS, this hardware supports 3000 pppoe sessions. In our infrastructure of more than 10000 clients we have installed 5 of these servers as cluster, since they are able to be clustered and of course works very good. I think it will be good to have the cluster feature in RouterOS, because I hope that with proper hardware, RouterOS can handle the same load as FinePoint BMS.

Regards.

Faton.

Too much users for a non professional PPPoE concentrator.

Maybe one day MKT kernel will handle it, but only when optimized for multi-core machines.

Split it on 5 MK servers (x86 machines with dual core processors).

Probably RB-1000 has enough power to route or NAT traffic for 5000 clientes, but not to act as a PPPoE concentrator.

even now with radius authentication you can support about 2000 pppoe on RB1000 if not more. but it was said - segmentation always helps to avoid big mob of angry users calling you at the same time and asking for internet and all of that just because conditioner in server room went "bork brok".

Shall we expect in the future clustering feature in RouterOS.

Faton

do you have an example of already existing implementation of clustering?

We have tested over 2600 connections without issues.

This is really a concerned issue, alot of things have been suggested in this forum with hope that there will be put into consideration, but the more we suggest, the less we see. Clustering idea is a very good one compared to a single machine doing that much when this is not a voice terminator.

I remembered we also talk about Userman being able to handle time segmentation billing, but we keep having new versions by the day without significant changes.

Why not take your time to put all these together in order to save us from daily upgrape?

We are already used to MT and will want to stick to it.

I tried RB1000 with ~500 PPPoE connections pushing about 20Mbps. CPU usage was over 80%. On x86 Intel Core2Quad server, same number of users and traffic is only 40% CPU, and that is with multi-core turned off due to it's instability. I would not use RB1000 for 2000 PPPoE sessions. Seems like it's CPU can't keep up.

cpu usage doesn't mean anything. RouterBOARD 1000 can easily run 10000s of clients provided you use a simple firewall and PCQ for the clients. It's all about optimization

Normis, can we expect clustering feature in the near future ?

cpu usage doesn't mean anything. RouterBOARD 1000 can easily run 10000s of clients provided you use a simple firewall and PCQ for the clients. It's all about optimization

As a simple firewall I believe but as PPPoE server I doubt. Every time a machine or Routerboard reaches more than 80% it increases latency and jitter. It´s impossible to respond the same way as 40% cpu usage.

it's been tested, you can believe me or not, but it's a fact

Again I will repeat myself, if Mikrotik developers will achieve to make RouterOS to have clustering feature, for 5000 clients I will not save to install RB1000 as many as it needs to operate at optimal mode. But, I still didn't get any answer.

Regards

Faton

we are not planning to make clustering

shame, it could have replace very expensive pppoe-servers like FinePOINT ServPoET.

it's been tested, you can believe me or not, but it's a fact

You tested it as a PPPoE server with 10.000 clients?

What was the clients bandwidth? How much traffic thru router?

Testing and real world are completely different situations. Sure you can fire up 10,000 sessions and say "well, they all logged in", but that's not even remotely the same thing as having 10,000 actual users using their internet access!

However, I feel that with a suitably large x86 PC with a sufficient number of network interfaces handling the load, there's no reason why MT shouldn't be able to handle 5,000 users.

I have no objective data to support my opinion, I have just been using MT for a long time and I feel that it's reasonable to expect that it would work with that many users as long as your configuration is sound and you aren't trying to do it on something that can't handle that level of traffic.

Hi,

I have used MT with 3000+ PPPoE Sessions. It was working fine. A Local Cable Internet System with almost 15000 Clients

I was using it only for PPPoE Authentication purpose not for any other. My setup was Like This

INTERNET < -- > LINUX < -- > MT < -- > CLIENTS

It was working. I was managing every thing related to Firewall and Proxy from Linux, MT was only used for PPPoE and QoS.

Hi

Currently using MT 2.9.51 on x86 (Intel 1U server SR1425BK1 - 3.2ghz / 10gig IDE HDD / 2048ram)
PPPoE connections around 3300, CPU runs about 65% during peak client usage. We have 4 of these controlling a segmented network of about 150000 clients. Systems have been rock solid for over a year.

Decided a few weeks ago to try to get better performance by upgrading to MT 3.17 to take advantage of the 2048ram and dual core CPU. This lasted about 1 day, systems became very unstable with frequent lockups under load requiring hard reboot of system. Rolled back to 2.9.51 and smooth sailing again.

Hi,

yes its very true 2.9.5 is more stable then 3.X. I have recently done a setup of about 1000 clients on 2.9.5 and it is running very smooth, and me my self running about 100 clients on 3.20 and believe me its a headache.

this kind of complaining will do no good. if you have problems with pppoe on v3, make a supout.rif file and send to support. all of you.

Reefbum, sounds like multicore issue... you may try multi-cpu=no and still use advantage of 2048 Mb RAM =)

I too have been wondering about how much I can push on to a routerboard 1000 when it comes to PPPoE sessions. I am using radius to authenticate and putting rate-limiting on threw radius profiles- I've got 700 terminating now on a cisco 7206 hoping to move all of them over to the mikrotik with a bit of room to grow- just wondering in your opinions how much room there is if any? Do you find that memory or CPU is a bigger limiting factor? I see many people saying that 2.9.51 is the preferred platform to terminate PPPoE on, opinions on that? And does anyone have a good way to stress test a PPPoE terminator short of putting in in place?

No multicore issues in 3.17 and 3.18 works flawless on a IBM machine with a dual core Xeon 2 gigs of ram more than 300 4% CPU Use pppoe conections 3.20 MT queues seems dot not correctly no pppoe problems.

Your suggestion to set up a pppoe server is a 5000 client + QTree + Firewall