To add some information if useful. In our case there are 2 main “actions” performed as processing when http service is called that takes a lot of time (there are more but are negligible against these 2)
1. A query to an Diameter DRA (which for external reasons or customer reasons) takes an average of 150ms (and in some other cities of the deploys can take more because of the delay in the links) 2. A query to a rest webservice that is in AWS, not that bad as point 1 but still bad (~70ms) So, in the best of scenarios, we have ~225ms, maybe if the DRA load balances to another HSS we get ~180ms total call processing. That is bad, really bad. To probe the idea and before any major changes in any part, I started to keep DRA responses in a redis (we use dragonfly), it’s not consistent in terms of our call processing flow but my idea was to “remove” the delay of the diameter query (I run the query, keep the result in redis, when a new request arrives for the same value I use the cached value and send a new DRA query in other thread to avoid lock the current one and update the redis). With this, we removed the DRA query wait, so 220ms became 70/72ms. Instantly all cpu usage, all retransmissions, everything disappeared, cpu, mem went down drastically, etc etc. That’s why I mentioned ’wait time is the enemy’, http is not (maybe is not your best friend but not the enemy). If it works, there’s an analogy, you can try to improve aerodynamics, engine and whatever in a F1 car, but if the driver is heavy … Hope it helps > On Dec 19, 2024, at 5:18 PM, Alex Balashov via sr-users > <[email protected]> wrote: > > >> On Dec 19, 2024, at 1:06 pm, Calvin E. via sr-users >> <[email protected]> wrote: >> >> Consider scaling out instead of scaling up. Now that you know the >> apparent limit of a single node for your use case, put it behind a >> Kamailio dispatcher. > > On the other hand, you might consider scaling up first, since HTTP performs > so badly that there is lots of improvement to be squeezed out of optimising > that even a little. > > You might think of scaling out as a form of premature optimisation. :-) > > -- Alex > > -- > Alex Balashov > Principal Consultant > Evariste Systems LLC > Web: https://evaristesys.com > Tel: +1-706-510-6800 > > __________________________________________________________ > Kamailio - Users Mailing List - Non Commercial Discussions -- > [email protected] > To unsubscribe send an email to [email protected] > Important: keep the mailing list in the recipients, do not reply only to the > sender! __________________________________________________________ Kamailio - Users Mailing List - Non Commercial Discussions -- [email protected] To unsubscribe send an email to [email protected] Important: keep the mailing list in the recipients, do not reply only to the sender!
