Weird illustration of peculiar interactions :-}
So a few weeks ago, I finally got around to implementing (as recorded either here in amanda-users or in -hackers; I forget) a separate "Offsite" configuration that runs on Sundays, and does a full backup of everything (with "no-record"). Since I had been able to (just barely!) run amanda with a dumpcycle of 2 days, I figured that I should normally be able to get the full ("Offsite") backup in 2 tapes... so I defined a runtapes value of 3. Well, as it happened, one of the file systems I back up is hovering right around 12 GB. And as luck would have it, amanda would typically get about 30 - 35 GB on the first tape before this file system was ready to be taped. Then taper would try to write this backup image to tape, and eventually fail, only to re-start on the next tape (all as expected). But that would leave a bunch of wasted space on the end of that first tape, with the result that the Offsite backups would actually require all 3 tapes. Friday, I added another system to the set of backups -- one that is on our perimeter net (and thus, subject to firewall restrictions). I was able to use the "Firewall" entry in the FAQ-o-matic (thank you!), and then today, for the Offsite backup -- as I half-suspected -- the timing was such that the 12 GB file system wasn't ready to be taped until amanda had switched to the second tape already. So now, because I added another bunch of file systems (including one that is 3.5 GB used), the Offsite backup only used 2 tapes. Go figure. :-) (Tape chunking should address this rather nicely, of course) Anyway, I thought that some folks might find that vaguely amusing: >These dumps were to tapes E15, E16. >The next 3 tapes Amanda expects to used are: E17, E18, E19. > > >STATISTICS: > Total Full Daily > >Estimate Time (hrs:min)0:36 >Run Time (hrs:min)15:25 >Dump Time (hrs:min) 213:22 213:22 0:00 >Output Size (meg) 74162.474162.40.0 >Original Size (meg) 141433.4141433.40.0 >Avg Compressed Size (%)44.3 44.3-- >Filesystems Dumped 150150 0 >Avg Dump Rate (k/s)98.9 98.9-- > >Tape Time (hrs:min)3:49 3:49 0:00 >Tape Size (meg) 74167.174167.10.0 >Tape Used (%) 211.9 211.90.0 >Filesystems Taped 150150 0 >Avg Tp Write Rate (k/s) 5517.6 5517.6-- Cheers, david -- David Wolfskill [EMAIL PROTECTED] UNIX System Administrator Desk: 650/577-7158 TIE: 8/499-7158 Cell: 650/759-0823 I need help: http://www.whistle.com/employment/employ-engg.html#K030391 >From [EMAIL PROTECTED] Sun Jan 14 17:20:37 2001 >Date: Sun, 14 Jan 2001 17:18:37 -0800 (PST) >From: System Operator <[EMAIL PROTECTED]> >To: [EMAIL PROTECTED] >Subject: Whistle Communications AMANDA MAIL REPORT FOR January 14, 2001 >
Re: Weird illustration of peculiar interactions :-}
On Sun, 14 Jan 2001, David Wolfskill wrote: > Well, as it happened, one of the file systems I back up is hovering right > around 12 GB. And as luck would have it, amanda would typically get about > 30 - 35 GB on the first tape before this file system was ready to be taped. > Then taper would try to write this backup image to tape, and eventually > fail, only to re-start on the next tape (all as expected). But that would > leave a bunch of wasted space on the end of that first tape, with the > result that the Offsite backups would actually require all 3 tapes. I implemented some changes in the driver that cause it to gather dumps until a certain threshold is reached. Afterwards it will always write the biggest dump still fitting on the tape. This works quite nicely for me and improves tape utilization a lot. Unfortunately it also increases the total dump time a bit, if your tape is slow. I haven't released it yet, because I implemented it in Amanda 2.4.1p1 and didn't come around to porting it to 2.4.2. But if you like I can post the patches for 2.4.1p1. Greetings, Martin Martin Apel, Dipl.-Inform.t e c m a t h A G Group Manager Software Development Human Solutions Division phone +49 (0)6301 606-300Sauerwiesen 2, 67661 Kaiserslautern fax +49 (0)6301 606-309Germany [EMAIL PROTECTED] http://www.tecmath.com
Re: Weird illustration of peculiar interactions :-}
On Mon, 15 Jan 2001, Martin Apel wrote: >I implemented some changes in the driver that cause it to gather dumps >until a certain threshold is reached. Afterwards it will always write >the biggest dump still fitting on the tape. This works quite nicely for me >and improves tape utilization a lot. Unfortunately it also increases the >total dump time a bit, if your tape is slow. >I haven't released it yet, because I implemented it in Amanda 2.4.1p1 >and didn't come around to porting it to 2.4.2. >But if you like I can post the patches for 2.4.1p1. I have a perl script which will go through my holding disk and spit out a list of backup sets to select to best pack tapes. here's an example: [amanda@joi amanda]$ pack -C OffSite 138530/140906 ( 98%) /home/amanda/mnt/holdingdisk/OffSite/20010106 /home/amanda/mnt/holdingdisk/OffSite/20010114 116784/140906 ( 82%) /home/amanda/mnt/holdingdisk/OffSite/20010107 /home/amanda/mnt/holdingdisk/OffSite/20010108 114088/140906 ( 80%) /home/amanda/mnt/holdingdisk/OffSite/20010111 99327/140906 ( 70%) /home/amanda/mnt/holdingdisk/OffSite/20010112 Nothing left to pack! I amflush these to tapes to send to offsite storage. I've already done two tapes from this batch, I have four left to do. -- Joi Ellis [EMAIL PROTECTED], http://www.visi.com/~gyles19/
Re: Weird illustration of peculiar interactions :-}
On Wed, 17 Jan 2001, Joi Ellis wrote: > I have a perl script which will go through my holding disk and spit > out a list of backup sets to select to best pack tapes. > > here's an example: > > [amanda@joi amanda]$ pack -C OffSite > 138530/140906 ( 98%) > /home/amanda/mnt/holdingdisk/OffSite/20010106 > /home/amanda/mnt/holdingdisk/OffSite/20010114 > > 116784/140906 ( 82%) > /home/amanda/mnt/holdingdisk/OffSite/20010107 > /home/amanda/mnt/holdingdisk/OffSite/20010108 > > 114088/140906 ( 80%) > /home/amanda/mnt/holdingdisk/OffSite/20010111 > > 99327/140906 ( 70%) > /home/amanda/mnt/holdingdisk/OffSite/20010112 > > Nothing left to pack! > > I amflush these to tapes to send to offsite storage. > I've already done two tapes from this batch, I have four > left to do. That's a nice idea, but I have more data to back up than fits on the holding disk, so I have to flush some dumps to tape in order to dump all filesystems completely. Martin Martin Apel, Dipl.-Inform.t e c m a t h A G Group Manager Software Development Human Solutions Division phone +49 (0)6301 606-300Sauerwiesen 2, 67661 Kaiserslautern fax +49 (0)6301 606-309Germany [EMAIL PROTECTED] http://www.tecmath.com
Re: Weird illustration of peculiar interactions :-}
On Thu, 18 Jan 2001, Martin Apel wrote: >On Wed, 17 Jan 2001, Joi Ellis wrote: > >> >> I amflush these to tapes to send to offsite storage. >> I've already done two tapes from this batch, I have four >> left to do. > >That's a nice idea, but I have more data to back up than fits on the >holding disk, so I have to flush some dumps to tape in order to dump >all filesystems completely. > >Martin So? I was trying to point out that simply selecting the biggest dump may not give you the best packing. Often, the few tapes contain four or five smaller dumps and can obtain a 99.8% usage rate. The total amount of data to be backed up has no effect, since you can only flush what's on the holding disk. -- Joi Ellis [EMAIL PROTECTED], http://www.visi.com/~gyles19/
Re: Weird illustration of peculiar interactions :-}
On Thu, 18 Jan 2001, Joi Ellis wrote: > On Thu, 18 Jan 2001, Martin Apel wrote: > > >On Wed, 17 Jan 2001, Joi Ellis wrote: > > > >> > >> I amflush these to tapes to send to offsite storage. > >> I've already done two tapes from this batch, I have four > >> left to do. > > > >That's a nice idea, but I have more data to back up than fits on the > >holding disk, so I have to flush some dumps to tape in order to dump > >all filesystems completely. > > > >Martin > > So? I was trying to point out that simply selecting the biggest > dump may not give you the best packing. Often, the few tapes contain > four or five smaller dumps and can obtain a 99.8% usage rate. > > The total amount of data to be backed up has no effect, since you can > only flush what's on the holding disk. Yes, you are right. You might achieve a better packing by a more intelligent algorithm. But after you wrote out the first dump, things might have changed already because another dump has been finished in the meantime. I used this scheme for a few months now and it gives me usage ratios close to 100 % for all but the last tape. But this only works if you have a good mix of file system sizes. Greetings, Martin
Re: Weird illustration of peculiar interactions :-}
Martin Apel wrote: > > > > > So? I was trying to point out that simply selecting the biggest > > dump may not give you the best packing. Often, the few tapes contain > > four or five smaller dumps and can obtain a 99.8% usage rate. > > ... > Yes, you are right. You might achieve a better packing by a more intelligent > algorithm. I dont know if you noticed it, but you are talking about the famous "bin packing problem" in combinatorics. Just search the web for "bin packing" and you will find quite a few algorithms and further literature on this vast subject (even AMANDA uses one, according to some old papers). -- Regards Chris Karakas DonĀ“t waste your cpu time - crack rc5: http://www.distributed.net
Re: Weird illustration of peculiar interactions :-}
On Fri, 19 Jan 2001, Chris Karakas wrote: > Martin Apel wrote: > > > > > > > > So? I was trying to point out that simply selecting the biggest > > > dump may not give you the best packing. Often, the few tapes contain > > > four or five smaller dumps and can obtain a 99.8% usage rate. > > > > ... > > > Yes, you are right. You might achieve a better packing by a more intelligent > > algorithm. > > I dont know if you noticed it, but you are talking about the famous "bin > packing problem" in combinatorics. Just search the web for "bin packing" > and you will find quite a few algorithms and further literature on this > vast subject (even AMANDA uses one, according to some old papers). Yes, I knew that this is a known problem. But most of these algorithms are designed to work with full knowledge, i.e. they assume they know all sizes of all dumps. In Amanda's case this is not true, because after you wrote out a first dump you can make a new decision with extended information. After all, since I have rather good experience with my simple approach there's not really a need for a better algorithm. Regards, Martin Martin Apel, Dipl.-Inform.t e c m a t h A G Group Manager Software Development Human Solutions Division phone +49 (0)6301 606-300Sauerwiesen 2, 67661 Kaiserslautern fax +49 (0)6301 606-309Germany [EMAIL PROTECTED] http://www.tecmath.com