Re: Backup is too slow - configuration?
On Friday 24 January 2003 05:31, Harri Haataja wrote: >On Wed, Jan 22, 2003 at 01:55:32PM -0500, Gene Heskett wrote: >> On Wednesday 22 January 2003 09:05, hochenaw wrote: >> >we have an HP Surestore dlt vs80e and use hp dlt IV tapes >> > (capacity 40/80GB) under linux. > >... > >> >How can i activate software or hardware compression (cant find >> > an entry in amanda.conf, just Client or Servers best in >> > dumptypes-file)? >> >> Hardware depends on the os in use. Some os's have a choice of >> compressed or uncompressed drives in the device list and will >> turn the drives compression on and off according to the >> devicename you used to address it. Linux does not however, so >> one must find the switch setting on the drive itself that turns >> this on/off. > >... > >> Off is the generally recommended hardware setting for use with >> amanda because if the machines have the horsepower to do their >> own compressing, they can often beat the hardware compression by >> quite useable amounts, thereby putting more on the tape than the >> hardware compressor can. > >... > >I use this in my /etc/modules.conf on Linux: >post-install st mt datcomp off > >Which uses mt to switch off compression every time the module > loader autoloads the scsi tape driver. Unforch, this also doesn't actually do any good if the tape itself has ever been written to with the compressor turned on. The tape is scanned by the drive to see what it is, and if it finds the compression turned on in the tapes hidden header, the drive will turn it on despite your wishes. To convert a once written in compressed mode tape to an uncompressed tape takes a bit of fooling around with dd and mt. First, rewind the tape dd the amanda header out to a scratch file rewind the tape again use mt to turn off all the compression stuffs, thre are about 3 ways. use dd to write about 10 megs worth of /dev/zero to /dev/st0 dd the amanda header back to the tape which was rewound by your use of /dev/st0 above. This 10 meg write forces the drive to flush its buffers to the media while the compression is turned off, thereby resetting that flag in its hidden header. -- Cheers, Gene AMD K6-III@500mhz 320M Athlon1600XP@1400mhz 512M 99.22% setiathome rank, not too shabby for a WV hillbilly
Re: Backup is too slow - configuration?
Harri Haataja on Fri 24/01 12:31 +0200: > > Hardware depends on the os in use. Some os's have a choice of > > compressed or uncompressed drives in the device list and will turn > > the drives compression on and off according to the devicename you > > used to address it. Linux does not however, so one must find the > > switch setting on the drive itself that turns this on/off. > > I use this in my /etc/modules.conf on Linux: post-install st mt > datcomp off > > Which uses mt to switch off compression every time the module loader > autoloads the scsi tape driver. Isn't this what `stinit' is for? This associates the device files with arbitrary modes in the driver. It just doesn't hardcode them in the device file name, but lets you associate it at runtime (usually at boot time).
Re: Backup is too slow - configuration?
On Wed, Jan 22, 2003 at 01:55:32PM -0500, Gene Heskett wrote: > On Wednesday 22 January 2003 09:05, hochenaw wrote: > >we have an HP Surestore dlt vs80e and use hp dlt IV tapes (capacity > >40/80GB) under linux. ... > >How can i activate software or hardware compression (cant find an > >entry in amanda.conf, just Client or Servers best in dumptypes-file)? > > Hardware depends on the os in use. Some os's have a choice of > compressed or uncompressed drives in the device list and will turn the > drives compression on and off according to the devicename you used to > address it. Linux does not however, so one must find the switch > setting on the drive itself that turns this on/off. ... > Off is the generally recommended hardware setting for use with amanda > because if the machines have the horsepower to do their own > compressing, they can often beat the hardware compression by quite > useable amounts, thereby putting more on the tape than the hardware > compressor can. ... I use this in my /etc/modules.conf on Linux: post-install st mt datcomp off Which uses mt to switch off compression every time the module loader autoloads the scsi tape driver. -- Ah, young webmaster... Java leads to Shockwave. Shockwave leads to Realaudio. And Realaudio leads to suffering. -- Peter da Silva
Re: Backup is too slow - configuration?
On Wednesday 22 January 2003 09:05, hochenaw wrote: >Hi, > >we have an HP Surestore dlt vs80e and use hp dlt IV tapes > (capacity 40/80GB) under linux. >But the backup is to slow, we should have a backup rate of 10.8 GB >(native)/20.6 (comprimized), but we do not have :-( >So i have some questions, to which i did not find answers until > now! How did you arive at those figures? >In the file tapelist we have following entry (i dont know why): >- >define tapetype HP-DLT { > comment "HP SureStore DLT" > length 80 gbytes # conservative estimate > filemark 1 byte# should work given above > speed 30 mbytes# even more, but this isn't used > in amanda >} >- 80 gigs would not be what our experience would call a "conservative" estimate. Probably optimistic by several gigs. >length in comprimized (80) or uncommprimized (40) modus? >What does filemark mean? Each drive has its own method of laying down a file seperator marker, with some drives not even needing it. I don't know about the DLT though. Amtapetype will give you this data too. >speed in mbytes per minutes or ...? Does it have any consequence > if i insert to much/less speed? Speed is (I think) how fast the drive can stream data in per second. I think thats only for use by amanda in estimating how long it will take to write the file being instantly processed in an amstatus /config/ report. I don't actually know if this is used for other purposes too. To get a much better idea of the tapes actual capacity, the latest snapshots have an 'amtapetype" command, older versions call it just tapetype. Run this against a tape you can afford to lose the data on, and it will give you a much better description that you can incorproate into your tapetype list in amanda.conf. >How can i activate software or hardware compression (cant find an >entry in amanda.conf, just Client or Servers best in > dumptypes-file)? Hardware depends on the os in use. Some os's have a choice of compressed or uncompressed drives in the device list and will turn the drives compression on and off according to the devicename you used to address it. Linux does not however, so one must find the switch setting on the drive itself that turns this on/off. Software is controlled by the entry's in the 'dumptype' you choose. Off is the generally recommended hardware setting for use with amanda because if the machines have the horsepower to do their own compressing, they can often beat the hardware compression by quite useable amounts, thereby putting more on the tape than the hardware compressor can. Here, the difference is about 10%. Even though I do not compress every entry in the disklist, the overall average output size delivered to the media is about 40% of what is on the drives. Thats a very usefull advantage over the 'hardware' model IMO. If you are setting up a client/server lashup, you can also have the clients do the compressing, which has the added advantage of lowering the network bandwidth to move the files to the server, and thereby speeding things up a bit since several clients can be compressing at once whereas the server can only do one file at a time. Also, with the hardware turned off, amanda can do a much better job of estimating what will fit on the tape because amanda counts bytes sent to the drive regardless of the content of that byte. With hardware on, you can almost double the size in the tape type, but then amanda has no way of knowing how much the data is actually compressed so you have to reduce this double by a fudge factor. When running the amtapetype (or tapetype) program, if the hardware compressor is on, it will often give a very conservative value for the tape size and it won't be anywhere near the size in the makers propaganda. Thats because this utility uses /dev/urandom as the data source, and the data from /dev/urandom will often expand by 10-20% in going thru a hardware compressor. >I have differnt directories to backup, and i want to pack them in >_one_ big package, so i specify a chunk size of 20GB, but Amanda >creates for each entry in the disklist one package/file. Thats doable, but a 1 gig chunk size will prevent you from running into filesystem limits, often at the 2 gig mark. Trying to pack it all into one big file has no real value as amdump simply starts at the head of the tape, writing these disklist made files until it is done. This way if you were to need to recover say, /etc/X11/XF86Config, amanda will know it can skip all the other files on the tape and go directly to the file containing the one you want. One big file also means that since amanda cannot span to a second tape with any given disklist entry, if its too big for the tape, it will restart that disklist entry on the next tape until its either used all the tapes,
Re: Backup is too slow - configuration?
On Wed, 22 Jan 2003 at 3:05pm, hochenaw wrote > But the backup is to slow, we should have a backup rate of 10.8 GB > (native)/20.6 (comprimized), but we do not have :-( GB is not a rate, it's a size. I assume you mean GB/hour? > In the file tapelist we have following entry (i dont know why): > - > define tapetype HP-DLT { > comment "HP SureStore DLT" > length 80 gbytes # conservative estimate Err, no, actually that's a very optimistic estimate. Very rarely does hardware compression get you 2:1 as the manufacturers claim. And, if you're using software compression, you need to *not* use hardware compression and thus tell amanda the native capacity. > filemark 1 byte# should work given above > speed 30 mbytes# even more, but this isn't used in > amanda > } > - > length in comprimized (80) or uncommprimized (40) modus? Again, it depends on whether you use software or hardware compression. > What does filemark mean? The amount of space the tape drive uses when it writes an EOF sequence. > speed in mbytes per minutes or ...? Does it have any consequence if i > insert to much/less speed? Nope. > How can i activate software or hardware compression (cant find an > entry in amanda.conf, just Client or Servers best in dumptypes-file)? Hardware compression is controlled either via dip switches on the drive itself or software (e.g., 'mt compression 1' on Linux with some drives). Software compression is controlled in the dumptype. > > I have differnt directories to backup, and i want to pack them in > _one_ big package, so i specify a chunk size of 20GB, but Amanda > creates for each entry in the disklist one package/file. Yes, that's how it works. Each disk list entry has to be a separate file. > A hardware question: If i have many packages to flush out to tape, are > they written at once so that the tape comes into the streaming mode or > does the tape have to be stopped and restartet for each package newly? Amanda does it's best to keep the tape streaming. > What happens if i delete some files in the holding disk manually? > Does it have consequences and the database is corrupted? Amanda will think that those directories got backed up, but you don't have 'em any more. -- Joshua Baker-LePain Department of Biomedical Engineering Duke University