Yudi.
I use kannel 1.4.1stable in a production environment with vmware, and because this, i "had" chose to use an sqlbox for storage my DLRs(all types supported, like delivery report to mobile, delivery report to smsc, read report, etc.). With this, despite the kannel CRASHs, the DLRs are recovered from database and the works straight ahead. With sqlbox the leaks are minor, but still there. Like i said on later email, without vmware, the leaks consume the memory, but don't break the bearerbox, despite the performance's down. I think this is a problem in a memory management(more precisely "swap management") from VMWare. Finally, if you test the gateway with a vmware(probably a vmware workstation version) in a notebook, be aware that the performance and benchmarks are just approximate(because the performance of notebook, and because the performance of vmware workstation version). I already works benchmarks and crash tests in this way(it's so practical! :) ), but like my production is on a VMWare Server environment, with robust host servers, the difference appears. With my experience, i strongly suggest physical servers for kannel production environments. I test some others SMS gateways, but anyone is so mature and so robust like kannel. + Points to Developers ! Regards. Allison Almeida.