Very clear your information, appreciated, I'm being a little paranoid hahaha, thanks!!! On Mon, May 28, 2012 at 5:48 PM, Amos Jeffries <squid3@xxxxxxxxxxxxx> wrote: > On 29.05.2012 09:45, Beto Moreno wrote: >> >> Sorry, I didn't finish this post. >> >> 1 question, I will setup a cache server with squid 2.7.x, the server >> will have 8GB of ram, the biggest service will be squid, I'm thinking >> in 4GB for squid, 4GB other services. >> I have seen that squid use part of the memory to save objects and some >> times he return the cache from memory or disk, memory is faster we >> know this, them this is one reason >> that I want to give more memory to squid. >> >> Is what I understand, them if I'm right, what happen once U reboot my >> server? >> >> What does squid to the object in memory? > > > Squid tries to save what it can and terminate cleanly in the time allowed > (shutdown_timeout). > >> Do I lost them? > > > There is a very important concept to keep in mind here which is often > overlooked: > > *** Cache is temporary storage ***. Emphasis on "temporary". It is > Constantly renewed from network sources. > > As a result no content is ever "lost". Even should you completely erase the > cache contents and restart from a clean-slate, everything is still available > from the relevant origins/master server somewhere out on the network. > > > > >> Or haven't understand how squid use the memory? > > > You seem to understand the usage okay. But you are fixated a bit on hoarding > data. > > The only side-effect of rebooting the server is a short "full" outage of > service, and a short period of slightly slower service as the caches are > re-filled. > > This may sound bad at first glance, but it is a tradeoff between network and > disk re-builds. Even if everything was saved to disk across the reboot there > is still a slow period from disk I/O latency and CPU to re-populate the > memory indexes. Versus network lag of fetching a clean copy. > > Squid has automatic client new-connection damping/buffering to cope with > both these periods and DoS situations without actually loosing clients > requests. > > Amos >