EoIP

pcmd · Sun Dec 05, 2004 9:17 pm

I have linked two remote networks with EoIP over PPTP to bridge IPX traffic. It works but is very sloooowwww. One site has a 2Mb cable connection, the other a 384Kbs DSL. I've tweaked MTU and MRU settings without making any difference in speed. At this time encryption is not set up on the PPTP pipe (encoding is blank under pptp-server server). Any suggestions

mag · Sun Dec 05, 2004 10:35 pm

what router os version is in use?
what means slow in kbyte/s? 384 kbps will give you a throughput of around 20 kbyte/s.

regards.
matthias

pcmd · Sun Dec 05, 2004 11:08 pm

OS 2.8.19. What is the best way to quantify my through-put. An application I use is fully launched in 30 sec. on the LAN, 3min. 30 sec. to launch over the EoIP.

dwright · Mon Dec 06, 2004 5:46 am

try an ftp transfer over the the tunnel or use the bandwidth test server in mikrotik

http://www.mikrotik.com/docs/ros/2.8/to ... st.content

Dan

wjw · Mon Dec 06, 2004 7:32 am

OS 2.8.19. What is the best way to quantify my through-put. An application I use is fully launched in 30 sec. on the LAN, 3min. 30 sec. to launch over the EoIP.

Your limiting factor is the 384kbps DSL line so... if you have 10mbps ethernet, the program shoud load in

10,000 / 384 = 26 times the time it takes on your LAN. Based on it being the data rate thats slowing it down. So it's not doing too bad...

mp3turbo2 · Mon Dec 06, 2004 7:52 am

hi pcmd,

I can confirm - my experiences with tunnels are very similar. We were able to push around 80kB/s = 640kbit/s on 34Mbit e1/t1 line... which without tunnels passes >32Mbit/s...

btw, what is the FASTEST tunnel in MT? I thought IPIP should be the fastest because of its simplicity and lack of encryption.

bye, mp3turbo.

mag · Mon Dec 06, 2004 9:24 am

no problems with tunnels anymore. we're using a lot of EoIP and PPTP tunnels and do not have speed problems (up to several mbit/s per tunnel).

i haven't tested ipsec yet, but using tunnels has become a very important MT feature for us.

regards.
matthias

mip · Mon Dec 06, 2004 3:56 pm

Is that 384kbps dsl lien simmetric? 'cause here all dsl lines are assimetric. So if you have a 384kbps dsl, the upload will be only 64kbps, os anout 8kbyte/s.
Same about tunnels: we use eoip and no speed limit, just fine transmission.

pcmd · Mon Dec 06, 2004 5:08 pm

DSL connection is symmetrical, fractional T1

Tested with b-width tester, udp, 1500 packet size, both directions, results as follows:
Remote (dsl) to local (cable)2 minute test avg.
EoIP: 327/290kbs
PPTP: 324/285kbps
Public address: 349/323kbs

FTP'ed a 1778kb file to the router avg. 50 seconds on all three addresses.

Results testing from local (cable) to remote (dsl) were strange, tx kept dropping to zero then jumping back up:
EoIP:262/337kbps (0/256, 200/256,0/300, 287/300, 0/330, 320/330, etc.)
PPTP:281/331kbps (tx again would jump up and down from zero similiar to above)
Public address: 332/370kbps (again tx would jump as above)

Same results FTP'ing a file as above approx. 50 seconds to transfer a 1778kb file.

mag · Mon Dec 06, 2004 5:38 pm

DSL connection is symmetrical, fractional T1

hmm, t1 is not dsl, i guess you got 384 kbps with symmetric dsl (hdsl, sdsl)?

Results testing from local (cable) to remote (dsl) were strange, tx kept dropping to zero then jumping back up:

i don't think it's the mt. i would check everything between modems and routers (speed, duplex mode). i would check too, wether there are retransmissions.

perhaps some traffice shaping occurs on the dsl-line. at least here some dsl-providers are playing arround with transparent shaping.

regards
matthias

pcmd · Tue Dec 07, 2004 12:50 am

ISP confirmed 384K SDSL, all addresses pass thru, no shaping, no NAT'ing. 100Mbs full duplex at the local connection (cable), Winbox showing 10Mbs half duplex at the remote location (DSL). I'll have to visit out there and check that out.

pcmd · Tue Dec 07, 2004 12:52 am

Best way to check for re-tranmissions?

mag · Tue Dec 07, 2004 9:53 am

Best way to check for re-tranmissions?

a manageable switch or protocol analyzer in the ethernet segment between modem and router and between router and pc.

regards.
matthias

jonbrewer · Tue Dec 07, 2004 12:56 pm

hi pcmd,

I can confirm - my experiences with tunnels are very similar. We were able to push around 80kB/s = 640kbit/s on 34Mbit e1/t1 line... which without tunnels passes >32Mbit/s...

btw, what is the FASTEST tunnel in MT? I thought IPIP should be the fastest because of its simplicity and lack of encryption.

bye, mp3turbo.

Most of our network uses tunnels - IPIP and EOIP for delivery of customer traffic and PPTP to connect our network over an IX. It helps keep things clean as we are delivering traffic for a number of different ISPs and we have wireless nodes on a number of networks connected to the IX.

With 400mhz Celeron (Compaq SFF with cheap Realtek cards) as the tunnel servers configured below:

wireless node -- tunnel server -- ix border (ix over metro area fibre) ix border -- tunnel server -- wireless node

tool bandwidth-test 10.16.0.132 protocol=tcp direction=both random-data=yes

status: running
duration: 30s
tx-current: 52.1Mbps
tx-10-second-average: 52Mbps
tx-total-average: 52.3Mbps
rx-current: 22.6Mbps
rx-10-second-average: 20.9Mbps
rx-total-average: 19.9Mbps

The asymmetry is likely due to the fact that one of the tunnel servers is plugged in to the same switch twice for a bad, ugly reason. Still the throughput is much better than you report!

mp3turbo2 · Wed Dec 08, 2004 1:52 pm

Jon,

that's one excellent performance! Will you post your configuration of IPIP or EOIP tunnels? Omit IP addresses and passwords and stuff like that, I'm interested in config of IPIP (it's very simple, but we haven't had success with it - performance was low) and mainly with MSS/MTU mangling - do you do any mangling at all? How?

thnx, mp3turbo.

jonbrewer · Mon Dec 13, 2004 1:25 am

Jon,

that's one excellent performance! Will you post your configuration of IPIP or EOIP tunnels? Omit IP addresses and passwords and stuff like that, I'm interested in config of IPIP (it's very simple, but we haven't had success with it - performance was low) and mainly with MSS/MTU mangling - do you do any mangling at all? How?

thnx, mp3turbo.

We did mangle to correct for problems with MTU but have stopped. If the customer has a problem with PMTUD, we just set the MTU on IPIP to 1500 and let the packets fragment or we use EoIP which lets the packets fragment. The ones with MTU of 1440 are also going through PPTP tunnels at some point before they get to the customer. If we could get Mikrotik to support Ethernet cards with large MTU, then there would be no fragmenting at all!

On one of our routers, IPIP looks like this:

0 R name="tunnel-to-1" mtu=1480 local-address=10.16.0.65 remote-address=10.16.0.10

1 R name="tunnel-to-2" mtu=1480 local-address=10.16.0.65 remote-address=10.16.0.25

2 R name="tunnel-to-3" mtu=1480 local-address=10.16.0.65 remote-address=10.16.0.74

3 R name="tunnel-to-4" mtu=1440 local-address=10.16.0.65 remote-address=10.16.0.206

4 R name="tunnel-to-5" mtu=1480 local-address=10.16.0.65 remote-address=10.16.0.78

5 R name="tunnel-to-6" mtu=1480 local-address=***.7.1.10 remote-address=***.7.1.12

6 R name="tunnel-to-7" mtu=1440 local-address=10.16.0.65 remote-address=10.16.0.150

7 R name="tunnel-to-8" mtu=1500 local-address=10.16.0.65 remote-address=10.16.0.82

On the same router EoIP looks like this:

0 R name="eoip-1" mtu=1500 mac-address=00:00:5E:80:00:02 arp=enabled remote-address=10.16.0.138
tunnel-id=1

1 R name="eoip-2" mtu=1500 mac-address=00:00:05:80:00:04 arp=enabled
remote-address=10.16.0.142 tunnel-id=2

2 R name="eoip-3" mtu=1500 mac-address=00:00:05:80:00:05 arp=enabled remote-address=10.16.0.150
tunnel-id=3

pcmd · Mon Dec 13, 2004 4:56 pm

No joy with this test yet, still slooowww. I asked MT tech support about hardware requirements but did not receive an answer. MT box at remote side P166, 64Mb RAM, 3COM 3c509 ISA public interface, 3COM 905-TX PCI Private interface; Local side P233 w/MMX (not PII), 128Mb RAM, RealTek 8139's on Public and Private interfaces. I've monitored resources on both machines and rarely see spikes above 10% on CPU usage. Any thoughts from someone who actually has EoIP over PPTP working OK?

sten · Mon Dec 13, 2004 5:20 pm

MT box at remote side P166, 64Mb RAM, 3COM 3c509 ISA public interface, 3COM 905-TX PCI Private interface; Local side P233 w/MMX (not PII), 128Mb RAM, RealTek 8139's on Public and Private interfaces.

3COM 3c509-TX PCI cards generate corrupt packets when certain TX hardware acceleration is used. (I don't know if RouterOS uses this, maybe you should check if there are corrupt packets coming out?).

3COM 3c509 ISA cards generate too many IRQ's pr packet to be used for PPS intensive situations. (In my opinion).

Have you checked if there are any duplex mismatches?

I get what i expect from EoIP, but i must add that if there are are any hardware problems (physical connection is dropping packets etc). It will manifest itself more than in a pure physical connection. (More encapsulated packets being dropped than without).

mag · Mon Dec 13, 2004 6:44 pm

the realtek nics are causing more cpu-load, which is the limiting factor already.

our pc staff told me getting some cheap intel 845/celeron 2 ghz boards with on-board intel-nic will give full throughput. after testing, better or less energy-consuming hardware might be choosen.

regards.
matthias

pcmd · Tue Dec 14, 2004 4:24 pm

No duplex mismatches.

So 3COM nics are junk, Realtek nics are junk and I need to build up a couple of $400 boxes just to test if my hardware is a problem?

Again, my cpu usage rarely spikes above 10%..spikes, not sustained.

Thanks to everyone for their opions in this matter.

mag · Tue Dec 14, 2004 6:00 pm

...
So 3COM nics are junk, Realtek nics are junk and I need to build up a couple of $400 boxes just to test if my hardware is a problem?
...

i wouldn't call them junk, but doing bandwith or performance tests it's always important to know about hardware issues.

regards.
matthias

Hammy · Tue Dec 14, 2004 8:25 pm

Well, those 3COM NICs you are referring to are quite old. I'd say a good Intel NIC comes first followed by a good 3COM NIC. Haven't heard much about NVidia NICs, or the new Intel ones integrated into the chipset, but I'd suspect those NICs to be good as well. As far as anything else? Garbage as far as I'm concerned.

pcmd · Tue Dec 14, 2004 9:17 pm

Setting up a pair of PIII 450's with Intel Pro 100 NIC's at both ends of the tunnel to see if there is any difference in thru-put. I'll post my results.

Thanks to all again!

pcmd · Wed Dec 15, 2004 1:34 am

Installed PIII 450's with Intel Pro 100 nics on all ethernet interfaces (4). Knocked 12 seconds off my application launch time. Guess the next step is to fatten the pipe.

EoIP

EoIP

Re: EoIP

EoIP

Re: EoIP

Test result

Re: Test result

EoIP

Re-Transmissions

Re: Re-Transmissions

EoIP

Re: EoIP

Re: EoIP

EoIP

Re: EoIP

EoIP

EoIP

Who is online