MRU allocation update

[Hal Murray]

The details of how it works are in the ntpq man page.  Search for maxage or 
minage.

A server will end up operating in one of several modes, depending on the 
ratio of traffic and MRU table size.

If the server has only a few clients, they each get a slot and they will 
stick around until ntpd is restarted.  The default mindelth is 600 slots.  
The default table size is bigger than that.

If you have a lot of clients and the table is big enough, the table will fill 
until some slots are older than maxage.  Those slots will get recycled.  The 
table will never get full.  There may be slots older than maxage.  They don't 
get recycled until needed.

If you have a lot of clients but the table isn't big enough, slots won't make 
it to maxage.  The oldest slot will get recycled when needed as long as it is 
older than minage.  The table will be full.  The oldest slot will be 
somewhere between minage and maxage.  On systems with bigger tables the 
oldest slot will be nearer to maxage.

If you have even more traffic, some traffic will go unmonitored.  The table 
will be full.  The oldest slot will be minage.  Slots will get recycled as 
soon as they are old enough.  Some traffic will not get monitored.

ntpq monstats has been updated to show counters and the age of the oldest 
slot.

Let's look at some numbers.  Slots are roughly 100 bytes, so you get a 
million slots with 100 megabytes of memory.  At 100 packets per second, 
assuming they all come from different addresses, it takes 10K seconds to fill 
up the table.  A host can't be very abusive if it doesn't send at least 10 
packets.  So if all hosts are potentially abusive, it will take over a day to 
fill the table.

I have some data.  I'll have to make some histograms.


-- 
These are my opinions.  I hate spam.