Is sticky table really FIFO ?

[michael.e.reid]

Hi, looking at my sticky table on the CSS I can see entries that have been in their for an extremely long time - 3810341 seconds = 44 days.

We have many services on the CSS but one service on the CSS has 14,000 logons per hours, with a 32K sticky table I would expect entries to get purged within the day ?

Can anyone explain this behaviour ?

cheers,

Mike

[Gilles Dufour]

Mike,

could you go in llama mode and do a

"sho sticky-stats "

How many entries do you have and how many available entries ?

Do you have the sticky-inact-timeout configured ?

Gilles.

[michael.e.reid]

Hi Gilles,

First thing, we actually have 288Mb of memory so we have 128K sticky entries, not 32K.

I just spoke with the server guys and they have a total amount of users of 100,000 so it is possible the entries stay in the table for a long time as the other services do not have so many users.

We do not have sticky-inact-timer set but we are looking to do this for this service to see if the load on the servers becomes more equal.

CONTENTSWITCH(debug)# show sticky-stats

Total number of available sticky entries is 0

Total number of used sticky entries is 131071

Total L3 sticky entries are 131071

Total L4 sticky entries are 0

Total SSL sticky entries are 0

Total WAP sticky entries are 0

Total number of SIPCID sticky entries is 0

One other question, if a sticky entry exisit for Client A going to Server A, then server A fails. Client A will get redirected to Server B. Will the original sticky entry be overwritten with this new one ? I would expect so.

cheers,

Mike

[michael.e.reid]

Gilles,

Yesterday I accessed a service on the CSS, and saw my sticky entry in the table. Today that entry has disappeared which suggests that since I used the loadbalancer 128K + entries have been written to the sticky table.

However, I still see entries with time elapsed of over 30 days ! If a flow is currently active, is it removed from the sticky table ?

cheers,

Mike

[Gilles Dufour]

Mike,

yes, the sticky entry is overriden and clientA will now point to serverB

Gilles.

[michael.e.reid]

Thanks Gilles,

Could you answer my last point ?

cheers,

Mike

[Gilles Dufour]

Mike,

can't explain why the entry for your pc was deleted. Except if it was already there before without you knowing it.

An entry could be there since a long time and show a small Time if it was being used recently.

Gilles.

[michael.e.reid]

Gilles,

No, I checked I had no entries in the table before I connected.

If a device is constantly connected via a TCP flow to the service, will it's sticky entry NOT be affected by FIFO.

i.e Is it only inactive flows that are affected ? Is there a default 'sticky-inact-timer' ?

cheers,

Mike