Re: Strange behavior over WAN

El vie., 11 de oct. de 2019 a la(s) 11:38, Albert Shih ([email protected])
escribió:

> Hi
>
>
> We got a postgresql server with a very small database ( ~few Mo) and do
> a extract with a ruby (sequel) client.
>
> If I do the extract on the same server the speed is correct (10-30 sec)
>
> If I do the extract on the same LAN the speed is correct. (10-30 sec)
>
> If I do the extract from the wan the speed goes from slow to ultra-slow
> (reach the TTL and sometime the connection are close) (1 minutes -> 8
> minutes)
>
> When this happen I try to see if they are some bandwith problem, and they
> don't have any bandwith problem ( ~10-20Mbytes/s 120-180Mbits/s), the size
> of the extraction (with tcpdump) are small (~2Mo).
>
> I also check the fragmentation of the tcp flow, when I'm on the wan they
> are x2 packets so I can understand they are slower than over the LAN but I
> can see how it's possible to go 50-60x slower.
>
> Any idea ? ....or solution ?
>
>
Hi there. Your issue is most probably latency.
This article will give you some pointers on the subject and what you can do
about it (change your data extraction method):
https://www.2ndquadrant.com/en/blog/postgresql-latency-pipelining-batching/