Justin Mason a écrit : > On Tue, Dec 16, 2008 at 23:35, Karsten Bräckelmann > <guent...@rudersport.de> wrote: >> On Tue, 2008-12-16 at 15:06 +0000, Justin Mason wrote: >> >>> actually, Bayes would be a good one to drop. If you also remove AWL, >>> and comment out both "loadplugin" lines, you will remove the need to >>> load the DB_File database module too. >>> >>> commenting plugins and removing rulesets is definitely the way to go. >>> you may be able to keep some net rules active -- trading long scan >>> times and latency for low memory. >> Why is that? Given the, uhm, quite low specs of that box, scanning a >> message will consume a few seconds CPU cycles anyway. Adding net tests, >> which don't require CPU but just "add" latency (well, background tasks), >> should be less of an issue the longer SA needs anyway. > > I think we're agreeing. ;) Removing plugins means lowering the per-process > memory footprint. I presume there's a very low messages/minute throughput > rate -- ie. it's a 1-person scanning server, or similar. In that situation, > you > can set the number of spamd children to a low value and limit concurrency. >
he can also reduce postfix concurrency to avoid accumulating a lot of mail. > BTW you definitely want to use spamd, rather than the "spamassassin" script. > It's good at sharing memory. > > Obviously I wouldn't use this as a scanning server for an office full of > people, > but for a 1-man home network I can't see why not... it'd certainly be a nice > reduction in power consumption. ;) and noise reduction as well. > > --j. > >> Luis, out of curiosity -- why do you want to run SA on that box anyway? >> Atom powered boxes with plenty of RAM are available for 200 bucks. That >> box essentially would be bored and running idle, while still having the >> resources to add almost whatever you feel like.
