On 15:56 Thu 31 Mar , Ciaran McCreesh wrote: > On Thu, 31 Mar 2005 05:32:39 -0500 Bill Roberts > <[EMAIL PROTECTED]> wrote: > | Compiling is not only cpu intensive, but involves a great deal of > | read/write. > > Which is why you make /var/tmp/portage a tmpfs mount. Or at least we do > for sparc and mips stage and GRP builds... But then our build boxes have > eight or more gigs of RAM in them :) > > | Show me any other way you can easily get the following > | numbers from hdparm: > | > | /dev/md0: > | Timing cached reads: 2868 MB in 2.00 seconds = 1432.78 MB/sec > | Timing buffered disk reads: 410 MB in 3.01 seconds = 136.05 MB/sec > > You can quite easily beat that with a straight fibrechannel setup. > I've done the tmpfs thing before, but not on my current machine. Thanks for reminding me.
I got the WD raptors for free (well, I did a couple of builds for a friend). I never would have bought Western Digital, but I have been happily suprised. Anyone wants to donate a couple of those 15K rpm Seagates, I'll be glad to take them, but I can't really justify the expense. Point is, I think the premium paid for scsi is too high. In some 15 years of computing, I've only had one HD fail, that was an IBM Deskstar, aka DeathStar. I have hard drives that have been running essentially continuously since 1998 without problems. As is always the case, your mileage may vary. Must be those Scotch genes I inherited ;). Bill Roberts
pgpFPKBag3Qtv.pgp
Description: PGP signature