Hi there, i have a question for you guys

Can anyone point out about the ideal/optimal/suggested "queue size" on pfifo queue type for different devices (RB2011/RB3011/RB1100/CCR1009/CCR1036) and why?

Some devices require longuer queue size to avoid the queued packets to be dropped, but.. im not sure about the "correct" size for it or how to define it

The problem is that i dont know if setting it way too high might generate issues that im not aware of atm and if it is directly related to CPU or RAM

Anyway, any comments on the subject would be appreciated, thanks in advance.

Hi

If you don't want any dropped packets you shouldn't use queuing. Then, if bandwidth is large enough no packet will be dropped.
If bandwidth is limited on the other hand, if you want prio / QOS you will by definition be dropping packets.

The only advantage of memory (buffers for packets) is that you can equalise the traffic over time.
The large the queue the larger the equalisation period.
The faster the throughput the lower the equalisation period

With no memory -> just shaping.

A piece of information might have been missing

The user wasnt near to reach the queue limit (400mb), he was using about 300mb, and 50.000 packets per second, queue size was 70.000 on this case and still got queued packets (around 4k) and dropped if queue size is a little bit smaller

This particular case is with an RB1009 which is unable to get 300mb+ when the SIMPLE QUEUE is enabled. So, my other question is .. is there a PRACTICAL limit for simple queue at all or is something about RB1009 itself ? Any hint might help.

The problem was solved after moving everything to Queue Tree tho, but i want to know if there is any documentation about this scenario anywhere that might be considered useful

For order of magnitude
Make the amount of data sent at set max speed fit for about 200milli seconds

If you set a big enough buffer which depends exclusively on the workload from/to WAN, no packets will be dropped ever in that situation, but will consume more RAM and a little more CPU, as it has to track every new big chunk of buffer.
It's up to you that even QoS can almost offer guaranteed delivery, or else faster bandwidth.
The buffer is no magic number, as for example I have 20/4mpbs and my buffer is 4MB! (2 seconds delay at most with guaranteed delivery for good network management libraries); if I set 1MB, it atleast drops 1 packet in a whole day which, would not make any difference, but I got resources to use anyway (128MB RAM).
Even with that gigantic buffer, gaming ping is always 32ms or at most 46ms while downloading torrents full speed or Netflix, etc.