2nd try. With attachments. On Fri, 19 Mar 2004, Micha Feigin wrote: > > After I switched to ext3 on my installation partition (Acer TM 803, Debian > > stable/testing, based on Knoppix 3.2, heavily upgraded, kernel 2.4.24) I > > noticed a severe performance loss. Actually, the system suddenly felt > > quite sluggish (especially during start-up). I did some hdparm -t runs and > > discoverd that the transfer rate had dropped to around 12.9 MB/s from > > around 20 MB/s. That is a 30 % drop! I cross-checked with my desktop > > installation (basically the same setup, but of course very different > > hardware) and I only noticed (if at all) a drop in transfer rate of about > > 1 or 2 percent between ext3 and ext2. Also, the vfat partition on the > > laptop delivers the good old 20 MB/s as did the ext2 partition. On the > > desktop there is also only a small difference between vfat and ext3. > > > > Any ideas why this happened? And yes, I have enabled DMA and manually > > switched to UDMA 5 on all concerned drives. > > Should I go back to ext2? > > Well, actually I went back to ext2 and only to find out: that was not the reason for the disk to behave slowly. I have tried to locate the real reason, but with not much success so far. I also tried to follow Micha'a advice and did some testing: > > Try running top in parallel to doing some heavy disk activity but > something that is supposed to be low on memory (try moving a bunch of > files around) and check the cpu load (or anything else that can show > you cpu load). I used the bonnie++ benchmark program. The results (in html) are attached. The difference in sequential reads is apparent: 16771 K/sec for the Windows partition (vfat) as opposed to only 12176 K/sec for the Linux/ext2 partition. That's a 25% difference. Question1 : could this be related to the position of the partition on the disk? How could I verify this? Is the hda2 partition always directly adjacent to the hda1 partition? I have re-partitioned the disk sometime in the past (before I noticed the performance loss). Maybe the Linux partition got (unintentionally) moved to the end of the disk? Question2 : Why does this not happen on my Desktop? I also have a Windows/vfat partition (at the beginning of the disk) and a Linux/ext3 partition somewhere else on the disk, but there is *no* difference in disk I/O. I am at a loss. Any help? Cheers, Stefan (debian @ goessling . de)Title: Bonnie++ V1.03 Benchmark results
Sequential Output | Sequential Input | Random Seeks |
Sequential Create | Random Create | ||||||||||||||||||||||
Size:Chunk Size | Per Char | Block | Rewrite | Per Char | Block | Num Files | Create | Read | Delete | Create | Read | Delete | ||||||||||||||
K/sec | % CPU | K/sec | % CPU | K/sec | % CPU | K/sec | % CPU | K/sec | % CPU | / sec | % CPU | / sec | % CPU | / sec | % CPU | / sec | % CPU | / sec | % CPU | / sec | % CPU | / sec | % CPU | |||
entropy | 1G | 12297 | 48 | 12040 | 2 | 5118 | 0 | 10250 | 37 | 12176 | 1 | 101.5 | 0 | 16 | 4394 | 98 | +++++ | +++ | +++++ | +++ | 4705 | 99 | +++++ | +++ | 20103 | 100 |
Sequential Output | Sequential Input | Random Seeks |
Sequential Create | Random Create | ||||||||||||||||||||||
Size:Chunk Size | Per Char | Block | Rewrite | Per Char | Block | Num Files | Create | Read | Delete | Create | Read | Delete | ||||||||||||||
K/sec | % CPU | K/sec | % CPU | K/sec | % CPU | K/sec | % CPU | K/sec | % CPU | / sec | % CPU | / sec | % CPU | / sec | % CPU | / sec | % CPU | / sec | % CPU | / sec | % CPU | / sec | % CPU | |||
entropy | 1G | 8293 | 33 | 10220 | 2 | 6892 | 1 | 14403 | 54 | 16771 | 3 | 98.0 | 2 | 16 | 51 | 97 | 134 | 98 | 1517 | 100 | 56 | 88 | 167 | 98 | 244 | 99 |