Hi Manel, said that the word “optimised” with NodeJS don’t really fit, I would like you to quantify “a lot of small http connections”. Please report the number of new connections/sec and the concurrent number of connections you need
Luca > On 10 Feb 2015, at 21:09, Manel Domenech <[email protected]> wrote: > > Hi. > > We are developing a service that needs to make a lot of small http > connections to servers to check their health. > > Our idea is to have a few linux cheap servers to make these connections > around the world. We are thinking in cheap vps. These cheap vps have enough > bandwidth to support a lot of connections. But when we've tried with an > optimized lightweight nodejs application, it seems there are some limits for > the concurrency. CPU load and memory are in bounds, and also bandwidth, > checked with iptraf. We think the bottleneck comes from too many > interruptions in the NIC. > > We've looked for solutions and have found that a possible solution is to use > a tpc stack implementation out of the kernel: pf_ring, dpdk, netmap. But we > are not experts at that level of network stack (we are developers :)). > > The question is: it makes sense to keep looking this way or we are totally > wrong? In a cheap vps we haven't any control over the physical NIC, only the > virtual device. This matters? It's possible to use some of these drivers > (dpdk, pf_ring, netmap) over a virtual device without access to the physical > device? > > Thanks. > > _______________________________________________ > Ntop-misc mailing list > [email protected] > http://listgateway.unipi.it/mailman/listinfo/ntop-misc _______________________________________________ Ntop-misc mailing list [email protected] http://listgateway.unipi.it/mailman/listinfo/ntop-misc
