How to make GRE tunnel faster

Andrey Avdeev · ‎07-13-2017

Hi team!

Once again about GRE tunnel speed:

We have to ASRs and GRE tunnel between them (West and East sides). In general, we using this tunnel for our applications and services. But i think it's very bad idea using GRE tunnel for this kind of applications (for example sync between databases, rabbit mq services and etc.) because we have a huge delay ( 70+ ms) and sometimes we have issues with sync processes.

So my questions are: how to make this tunnel faster? I mean is it possible to make it faster? (there is no any other traffic at this tunnel)

we have very basic tunnel configuration and this configuration is similar from both sides:

interface Tunnel0
ip address 10.0.0.1 255.255.255.252
ip mtu 1400
ip tcp adjust-mss 1360
ip ospf cost 50
tunnel source a.a.a.a
tunnel destination b.b.b.b
tunnel path-mtu-discovery

I think the best way to make sync process faster - use MPLS transport between this sites.

Thank you!

Vasilii Mikhailovskii · ‎07-13-2017

Hello.

How do you know, that the tunnel causes any kind of extra delay or other issue?

Do you see any difference in RTT when you ping b.b.b.b or 10.0.0.2?

Andrey Avdeev · ‎07-14-2017

Hi Vasilii,

It's good to see you again ) (First time was at Cisco Connect Msk.)

in general, this issue solved. I mean there is no any issue with GRE tunnel configuration and yes you are right, I have no idea about any extra delay except RTT - 70 msec between sides. Moreover, i have read RFC 1323 and here you are: Throughput = TCPWindow / round-trip-delay = 65535 Bytes * 8/ .07 Sec = 936412.2857 Bytes/Sec = 7,4 Mbits/sec (very low speed for sensitive traffic)

Thank you!

Vasilii Mikhailovskii · ‎07-14-2017

Hello.

Thank you - Connect was a wonderful time.

You are correct about the bandwidth, but window size really depends on operating system (even version) and I think 500K window size may indicate use of legacy OS or client.

For example, Windows 7 introduced a lot of improvements in TCP/IP Microsoft implementation and let applications to use high-latency links much more efficiently.