6 replies to this topic

[riotw7]

Hi,

We have a project with N+1 components, each creating a log file in parallel.
On each boot the partition(NTFS) where the logs are created is checked with chkdsk /R.
These logs generate a lot of fragmentation as you can imagine, thus the chkdsk gets really slow.

What's the best method to avoid this fragmentation in your opinion?

TIA,

[WingedPanther]

Assuming you mean something like you have a bunch of threads each creating log files, why not have a thread to handle logging to a single log file. The other threads could send messages to the logging thread.

[riotw7]

Thanks... Thought of that... but this would generate two issues:

1) logs which come from otherwise rather autonomous entities would all be entangled and would require filtering before they would be readable (this is minor issue- we can make grepping a habit :))

2) we also have a Hypersonic database which writes it's data in a similar manner. And even though having just two files competing for clusters would statistically give a bit better fragmentation, the partition would still become rather fragmented over time...

[WingedPanther]

Can you have them log to different drives?

[riotw7]

Unfortunately not.
I can also say that increasing cluster size of the partition from 4K to 64K helps too little...

[WingedPanther]

What you may need to do is periodically copy the logs to a different folder (creating unfragmented files) and then delete the originals. When you're laying down two files on the same drive at the same time, you're just begging for fragmentation.

[riotw7]

Although this will limit the fragmentation significally in the long run, it will still allow there to be N+1 fragmented log files, which make chkdsk sessions to take 30 sec/minute...