> Actually, its not the case that More number of Clients are trying to get the > content from One of Server as Server Throughput shows equal load on all > interfaces of Server which is around 4 Gbps.
This contradicts (or I understood it differently) a bit what you wrote previously. But what I got was - if you shutdown (or disable) the "slow nginx" then the behavior shifts to the remainin. Which to me indicates that the balancing via URI doesn't result in even load to servers or at least with current software (nginx)/hardware configuration it leads to the same result which means that particular server instance isn't at fault. I reread your initial post and some other things don't seem right to me: - "We see that out of two server on which load is high i.e around 5" but later your write "Server is having 60 CPU Cores with 1.5 TB of RAM" - a 5 load on 60 core cpu machine means that server has only ~8% load which isn't very high or is it a typo? But you could look at the haproxy status page and see if there aren't big differences in currently active connections to backends and/or the "slow" backend doesn't have way more (completed) requests - as just plainly comparing network interface throughput doesn't always indicate the work the server needs to do - you could saturate the 4Gbps link by sending a single 100Gb file to a single client with ~500MB/s the same time another server could send 1Mb requests to 500 clients or 50000 clients each downloading 10kb but the resulting load could be way different. rr _______________________________________________ nginx mailing list [email protected] http://mailman.nginx.org/mailman/listinfo/nginx
