Amdump is terribly slow
When you believe that the performance of Amanda is not what your hardware is capable of, check these issues:
The plots generated by running amplot on the amdump.X files can quickly point to some bottlenecks in the setup.
- Are you using the holding disk? Large dumps may bypass the holdingdisk if that is not large enough. Typical for these DLE's is that dumper and taper speed are the same. You can also grep for 'PORT-DUMP' in the Amanda log files.
- Holdingdisk too small, resulting in less parallelisation?
- Is DMA enabled on the harddisks? What values are reported by "hdparm -tT /dev/md0"?
- Specified a wrong value for bandwidth in the interface definitions?
- Network problems, like wrong duplex setting on the interfaces? Look at the error counts on the interfaces in netstat -i.
- Use "compression fast" instead of "compression best", or switch server/client compression if one of those is too slow; disable software compression for slowest computers. Enable hardware compression (introducing a different set of problems!).
- Some heavy programs running at the same time as backups are being done?
- If estimates are too slow, consider switching to calcsize or server (statistics only) based estimates, see also Amdump: results missing.