TCP sessions terminated

[ambo]

vodacom3g (or anyone else willing to answer ):

Is there a reason that TCP sessions get killed by your inline appliances after just 10 minutes of inactivity? I can understand that there was a need to recover ports when we were still in NAT-land but those days are now behind us.

I would really not like to waste my precious cap on keep-alives... or more importantly - have to reconfigure numerous devices to send them

 

[vodacom3g]

vodacom3g (or anyone else willing to answer ):

Is there a reason that TCP sessions get killed by your inline appliances after just 10 minutes of inactivity? I can understand that there was a need to recover ports when we were still in NAT-land but those days are now behind us.

I would really not like to waste my precious cap on keep-alives... or more importantly - have to reconfigure numerous devices to send them

I guess you would build up massive tables of IP/Port pairs that might not be active anymore.

What's the minimum timeout you need?

 

[ambo]

I guess you would build up massive tables of IP/Port pairs that might not be active anymore.

You definitely need a timeout - otherwise the state table might explode

What's the minimum timeout you need?

Somewhere between 60 and 120 minutes would be sufficient for me.

Did a quick google and it appears that both Windows' and Linux's implementation of TCP have a built in keep-alive every 7200 seconds (default config). So maybe 121 minutes would be a good value

 

[morkhans]

This would be good. I often have my ssh sessions killed when they are idle.

 

[vodacom3g]

Looks like we cannot just up this timeout due to the way in which we count data.

But there might be a solution in a few weeks time.

 

[ginggs]

Looks like we cannot just up this timeout due to the way in which we count data.

But there might be a solution in a few weeks time.

You going to stop counting data?

 

[vodacom3g]

You going to stop counting data?

Yes, you'll be receiving a bill based on non-packet parameters. Billing will now be based on peak throughput seen over a 24 hour period, extrapolated for the next 60 day roll-over period.

Basically moving from a byte-count to bit/second model.