wlagmay@xxxxxxxxxxxxx said: > > Thanks to all, but to be more particular, Im going to use the machine with 8 > or > 12 Gig of physical memory for squid caching, and we all know that caching > consumes to much memory. Our objective actually is to cache the most popular > pages on the memory so that it will be faster to access by the clients. > > so far there are 3 ideas, 1st no swap dir at all, 2nd physical memory > multiply > by 2 or 3 and the 3rd one creating a swap with 512 MB to 1 Gig. On my > scenario, wherein im going to use the system for caching, which one is more > applicable? First, always use swap. The only common reason to choose not to use swap under Linux is on a workstation where the user believes VM pressure is causing unwanted latency for interactive operations. This scenario is easily resolvable with tunable nobs for the Linux 2.6 VM and its entirely unnecessary to avoid creating a swap partition. As you've discovered, with 12GB of RAM using a 2 or 3 multiply rule is hardly reasonable. Ideally you will pick a value based on testing your workload against the actual machine in question. Personally, I rarely allocate more than 4GB of swap and never less than 512M with today's large disks. If you're worried about performance, you can stripe swap over multiple disks or disk arrays. Granted, you should never heavily be in swap, but if the circumstance arises it allows you to recover in some fashion. It's the difference between a dead machine and a recoverable one. These days, I'd suggest a multiplier of simply 1, 0.5, or 0.25 for creating swap. Or did you _really_ need that 4-12GB of disk space for something more important than increased stability and availability of the machine in question? (If the answer is _yes_, buy _more_ disk.) _______________________________________________ LARTC mailing list LARTC@xxxxxxxxxxxxxxx http://mailman.ds9a.nl/cgi-bin/mailman/listinfo/lartc
