i don't have any official numbers, as i don't run any of the master mirrors, and never bothered to
test it out. i manage 2 here in canada though, and friend manages a 3rd, all 3 are running from
ramdisk, you're welcome to try them out for yourself if you wish. rsync4, rsync5, &
i noticed an instant jump in the speed when i moved my mirror from u160 scsi to ramdisk, it does
make sense if you think about it. you're serving roughly 100k tiny files to a bunch of people all at
once. that's a lot of seek time on any hard disk. i also sync the ramdisk every 30 mins directly to
ram. basically, it's a lot less disk thrashing. ;)
Georgi Georgiev wrote:
> maillog: 01/12/2004-16:30:40(-0700): Rob Baxter types
>>i think the point is that the data that gets cached can and does change
>>every 30 minutes, so caching doesn't really help.
> It does help. If data changes, it is first updated in the cache. New
> data is served directly out of the cache, while the hard disk is synced
> in the background. That's unless I am majorly mistaken about how caching
>>all 5 servers in the rsync.gentoo.org rotation are currently running
>>rsync out of ram. i think i can safely say they wouldn't be nearly as
>>fast as they are, running off a hard disk.
> Do you have some real numbers? I am not trying to doubt you too much. I
> am genuinely curious how big the speedup is and if it is worth the
> effort. All I wanted to point in my post is that the speedup is probably
> not that great, but I'd really like to see benchmark numbers if someone
> went to the trouble of doing it.
>>besides, ram is cheap
> That's pretty relative. $70 for 512MB is not cheap for me. One reason
> why my poor "server" is still running with 2x512MB *PC133*.
>>and has a small footprint (smaller blocks) using
>>ramdrive, why not use it.
> Well, you may be right. I guess I could try serving two trees and do
> some tests myself.
email@example.com mailing list