We did not enable squid cache, so I think memory is ok for our case, and we run squid servers (without cache, without cache cluster, just as forward proxy) more than 100 servers more than 1 years in AWS serveral regions with EC2 c3.xlarge on ubuntu 12.04. It was always running well.
Just after upgrade ubuntu 14.04, we found the memory usage increased. Server Spec: AWS EC2 c3.xlarge (4 Cores, 7.5GB Memory, 2 x 40 GB SSD) Before upgrade: 12.04: Memory usage is always less than 50% (3.5GB), will increase or decrease because traffic changes CPU is very low, same as Disk IO, B/W (In or Out) is 500Mb/s at most, is around 200Mb/s most of time. 14:04 Memory usage is about 80-90 % (nearly 7GB), will increase , but it decrease very slow, and always keeping more than 50% (3.5GB), CPU is very low, same as Disk IO, B/W (In or Out) is 500Mb/s at most, is around 200Mb/s most of time. I tested with squid-3.3.8 (ubuntu offical packages), and squid-3.5.11 on 12.04 and 14.04, I think it is most likely ubuntu related issue ? because same version, same configs, but different OS versions. 2015-11-30 11:37 GMT+08:00 Amos Jeffries <squ...@treenet.co.nz>: > On 30/11/2015 3:19 p.m., 风声 wrote: > > I try to use jemalloc, but from monitoring, there is no difference, > > > > I follow this guide: > > https://github.com/jemalloc/jemalloc/wiki/Getting-Started > > > > I used LD_PRELOAD to let squid use jemalloc. > > > > is there something wrong ? > > > > If I want to re-compile squid with jemalloc, how can i do that ? Can I > just > > use some FLAGS ? > > > > A number of build variables can be set on the command line to > ./configure. See the end of the "./configure --help" for the ones your > Squid supports. > > But, hold up a minute there. > > You have provided all sorts of snapshots of what the system thinks > memory usage is. But not told us anything about what values you expect > to see there, or what the machine has available. Just a blank statement > that its "high". > > I see that the *entire machine* is using ~8GB of RAM. For those of us > managing machines with 32-128 GB of RAM thats actually pretty low. For > machines with only a few hundred MB of RAM plugged in thats impossible > numbers. So please provide some context about the machiens limits. > > > Also, be aware that "Virtual Memory" is just that *virtual*. It is not > real in-use memory. Whenever fork() is usesd to spawn a child helper or > pworker process the current memory value of the parent process is > assigned to the child - regardless of whether it needs or uses it. > Effectively doubling the "Virtual Memory" numbers with every helper > started, even when the helper uses only a few KB. > > > I see from the netstat output that you have ~6318 sockets currently in > play. Thats up to ~1.6 GB of RAM just for connection buffering right there. > > Then whatever the cache requirements are. Reaching several GB of RAM in > actual usage is pretty easy. > > Then only after that is all added together multiply by however many > dozen or hundred helpers are being run, and thats what the virtual > memory numbers can say Squid is "using". Though in reality it is not > even close. > > Amos > > > _______________________________________________ > squid-users mailing list > squid-users@lists.squid-cache.org > http://lists.squid-cache.org/listinfo/squid-users >
_______________________________________________ squid-users mailing list squid-users@lists.squid-cache.org http://lists.squid-cache.org/listinfo/squid-users
