On Wed, Oct 27, 2010 at 09:42:34AM +0200, Thomas Marko wrote: > Jon, > > thank you for help! > > Am 25.10.10 18:31, schrieb Jon LaBadie: > > Amanda tries to get balance things so that each days dump will > > backup the same amount of data (actually, if compressed, the > > same amount of "tape" is more accurate). But it can't get there > > until it has a history of dumps and compression rates. Even > > with a history, if you have a relatively small number of DLEs > > and very different sizes, it can never completely balance > > things. Under these conditions amanda seems to full-dump the > > small DLEs frequently. > > So the best strategy would be to "balance" the DLEs to have the same > size.
Rather than saying it is the "best strategy" I'd say the best situation is all DLEs the same size. But we live in the real world (most of the time :) and that is not a reasonable expectation. > The issue that small DLEs are always full dumped is not that > problem, because we are talking about some Megs /etc and some files in > /root... The problem are the big ones. Right, you spend some time on the big problems and let amanda deal with the small ones. > Should I split the big DLEs up > into handsome parts, so that most of them are nearly the same size? Yes, but don't go overboard trying to get them exact. And sometimes it is impossible to split them up. An example is I have my Windows "machines" are virtual. Their "virtual filesystem" is a single file on its own filesystem. I have no way of spliting it. But if you have a large /home, it is easy to split it up based on the leading character of the user names. This has been covered several times on the list and may be in the wiki. > This also leads me to the question regarding the dimensions of > tape_length and splitsize and so on a asked Brian in a posting above: > > "Should I go for tapes with the length of the largest DLE (+ some GB > spare)? Or should the tapes be smaller and should I use tape_splitsize > and runtapes as mentioned in my last post?" Tape splitting has addressed a lot of old amanda "problems". I'd recommend using it. The only negative is recovery of dumps without amanda software is slightly more difficult. Normal recovery with amanda software is still fine. > > > There are parameters that control how many total simultaneous dumps max > > and how many simultaneous from one client max. I think I have them set > > to 4 and 1. > > You are talking about "inparallel" and "maxdumps", right? > > > Same as my setup. And I ran into the same situation, only one channel > > of SATA on my selected amanda server (a very old Pentium-3). I just > > spent $29 more for another controller so I could add two more disks. > > One isn't even in a cage, it is lying on the bottom of the case. > > Do you run a RAID over the disks? If yes, which level? If no, do you > allocate the vtapes over all disks? No raid, but I do have multiple disks collected under linux' LVM into single filesystems. One is the 2 x 1TB disks mentioned above and the second is a pair of 300GB external USB drives. I was thinking of including the USB drives and internal drives into a single volume group, but I was not confident the slow start-up time of the USB drives would not upset things. So I keep them is a separate VG. Total formatted capacity of the two VGs is 2.4TB. I divided this into 100 25GB vtapes. That is greater than the capacity, but I'm counting on the last tape of each run being only partially full. My experience says that is reasonable as the two VGs run between 80 and 90% full. Just as a single datapoint, my environment is 7 clients, 27 DLEs, 250GB of data currently, largest DLE is 37GB (24GB compressed). I run a dumpcycle of 2 wks, daily amdump runs. Runtapes is 4 but typically only 2, sometimes 3 are used each run. Currently 37 days of dumps are on the 100 vtapes, about 2.5 dumpcycles. HTH jl -- Jon H. LaBadie [email protected] JG Computing 12027 Creekbend Drive (703) 787-0884 Reston, VA 20194 (703) 787-0922 (fax)
