Wow. I asked for "opinions" and got a bunch. I really appreciate the time everyone took to provide me with this insight and information. In my personal experience, I have noticed that Solaris is brutally efficient at memory management and Solaris 8's "cyclical page cache" does well at reducing the need for large amounts of swap. Cyclical page cache replaced priority paging starting in Solaris 8. The general consensus seems to be 1xRAM up to 4gig of physical RAM and 4gig swap for any amount over 4gig of physical RAM. I received every range of answer from "none" to "2xRAM+whatever your apps can use". Clive McAdam gave the first reference to Sun's "official" recommendations, which while vague, gives a good starting point. The link is http://docs.sun.com/db/doc/817-6960/6mmah94ck?q=swapfs&a=view Between reading this and other information, it seems that anything over 4 gig will not be utilized, even in a crash situation, as the core dump will be essentially "swapped" to disk in /var/crash/`uname -n`/ (or your designated crashdump location) as it is created - provided that diskspace in the crash location is adequate. Kalyan Manchikanti gave an excellent link to a Sunsolve article with additional information (Sunsolve login required to access this link): http://sunsolve.sun.com/private-cgi/retrieve.pl?doc=infodoc%2F16987&zone_32=s wap%2A%20size%2A%20 Russell Page gave good insight on his reasoning behind his personal core dump strategy as well as an interesting historical note on how the 2xRAM rule got started: "Historically, the 2xRAM rule was an artifact of the virtual memory architecture on a VAX. Vaxen couldn't demand page off the file system, so loading a progam meant copying the image from the file system to the swap area, then paged it in from there. So on a VAX, a large chunk of the swap space was no longer available for swap, and so admins had to configure lots of swap." I also found some good information Googling. One page of interest: http://web.brandeis.edu/pages/view/Network/Solaris8CoreDumps Thanks go out to everyone who responded. While far too many responded to name everyone, the following people provided the most detailed and helpful answers: Clive McAdam Kalyan Manchikanti Russell Page Nicolas Dorfsman Thomas M. Payerle Jon Hudson -- Robert Kracke Southern Company, EMS Systems Administration Support@emss.com postmaster@emss.com -- ----- Original Message ----- From: Robert Kracke To: Sun Managers Sent: Thursday, August 12, 2004 9:06 AM Subject: Your opinion on swap size Recently, we have been getting more midrange systems where in the past we have settled for multiple small server systems (typically V480 and V880). In the past, I have used the simplistic 2xtotalRAM formula to determine a general baseline for swap. At present, the systems I am working with are starting in the V1280 range, typically have 8 or 12 processors and a minimum of 16gig RAM. I can't imagine a system would need 32gig of hard disk swap, but I may be wrong. I also realize that the intended use of the system plays into this formula (i.e. simple web server vs. raw number crunching vs. big databases). Can anyone point me to some basic guidelines or offer advice on an effective and efficient swap formula? Any assistance provided is appreciated and a summary of responses will follow. _______________________________________________ sunmanagers mailing list sunmanagers@sunmanagers.org http://www.sunmanagers.org/mailman/listinfo/sunmanagersReceived on Wed Aug 18 09:48:54 2004
This archive was generated by hypermail 2.1.8 : Thu Mar 03 2016 - 06:43:36 EST