Hi, today i started checking all my MT 2.9.28 boxes to try to understand a couple of things..
The first strange thing is that memory usage is abnormally different on some devices (not only between mine and the some other users, but also between mine).
I had to reboot one and noticed that the amount of free memory after reboot is normal (around 16 megs on a 32 megs RB).
On the other end, devices that lock up have low memory (3/4 megs) and devices that have problems with bgp sessions have around 7 megs..
This reminded me of one of “Changeip” answers about bgp crash with low memory..
i double checked, and in fact all the devices with problems have less than 10 to 7 megs of free memory.
Routing table is not that big, it’s around 700 prefixes or less (it’s aggregated, or better filtered since aggregation does not work) so that on the internal network, when possible, only /24 are announces.
Less than 2.9.28 devices do not show any relevant memory load, nor do 2.9.28 show any more memory free by flush the route table.
Maybe there’s some memory leak on BGP code?
I’m just guessing, but i need to know something =/ a week (a very unpleasant week) has passed, and so far there’s not even a sign from MT guys..
No, not visible or wanted at least.. on every device when it’s just started memory is fine.. after several hours memory drop to few megs free.
I.e:
You take the device and power it up.. it show around 18/20 megs free..
It imports routes, memory drops to 7/8 megs free, than goes to 15 about megs free.
I guess this is the routes import and select process..
The device keeps memory usage at this level for some times, than free memory drops.. If i turn off bgp and flush ALL routes then, no memory is gained (i.e. even with no routes load, free memory is still 2/4 megs)
The only way to “recover” memory is to reboot
I sent two supouts file of two devices with this memory usage (and bgp session crashed).
It’s pretty easy to show the problem, as it is consistent (all the devices, after some hours show this pattern).
It’s not something that happens randomly.
Bye,
Ricky
Oh and if you mean if there is any difference between 2.9.28 and earlier versions config, yes, there is.
2.9.28 uses one single instance for all peer,and has force-nexthop set to self.
2.9.2x has one instance for each peer, and force-nexthop option does not exist.