Re: GBDE performance on ZIP disks

Robert Watson, 28.10.03, 03:26h CET:

[...slow gbde encrypted ZIP disk...]
> How do things look performance-wise if you do a raw sector read comparison
> with dd at various blocksizes?  My recollection is that our msdos code

Here are a few numbers (for reading - the ones for writing don't really
differ):

[13:45] stefan_at_kyuzo> sudo dd if=/dev/da1 of=/dev/null bs=512 count=1000
1000+0 records in
1000+0 records out
512000 bytes transferred in 20.559991 secs (24903 bytes/sec)
[13:45] stefan_at_kyuzo> sudo dd if=/dev/da1 of=/dev/null bs=8192 count=1000
1000+0 records in
1000+0 records out
8192000 bytes transferred in 25.996819 secs (315115 bytes/sec)
[13:46] stefan_at_kyuzo> sudo dd if=/dev/da1 of=/dev/null bs=32768 count=1000
1000+0 records in
1000+0 records out
32768000 bytes transferred in 46.302189 secs (707699 bytes/sec)
[13:47] stefan_at_kyuzo> sudo dd if=/dev/da1 of=/dev/null bs=65536 count=1000
1000+0 records in
1000+0 records out
65536000 bytes transferred in 77.230242 secs (848579 bytes/sec)

> would benefit hugely from the addition of clustering support, but UFS2
> with a fragment size matching GBDE's notion shouldn't present the same
> problem...

The fragment size practically doesn't matter; I tried newfs's defaults
(block size 16384, frag size 2048) as well as a fragment size matching
gbde's sector size (-> block size 4096, frag size 512), and the time
difference for copying the mentioned 37 MB of data was <1 second.

Stefan