• Advertisement
Sign in to follow this  

Individual disk access on RAID 1

This topic is 4211 days old which is more than the 365 day threshold we allow for new replies. Please post a new topic.

If you intended to correct an error in the post then please contact us.

Recommended Posts

Is there a way to circumvent the fact that on a RAID 1 (mirroring) system a pair of hard drives are seen as one and, when writing a file, override the default "write in parallel mode" behaviour and instead write in a weird kind of striping mode by interleaving the written segments in the following way:
[start write]

[write]
HDD1  ----    ----    --
HDD2      ----    ----

[end write]

[synchronize]
When writing is finished, free up control from the output process, but spawn a background sychronization thread that copies the missing portions to the complementing hard drive, maintaining the RAID 1 parallelism. This would allow increasing write speeds two-fold on RAID 1 systems during periods when fast output to disk is the priority, but will require more time later on to do the synchronization. Could be very useful for some applications. I do realize that there are probably no ready means of accessing RAID 1 disks individually, especially if RAID is maintained in hardware; and if there is a way, it would most likely entail writing a device I/O driver. However, conceptually, this shouldn't bring about added risk if one of the drives fails while they are not synchronized since the only file that would be lost is the file that is being written.

Share this post


Link to post
Share on other sites
Advertisement
Considering that hardware can make RAID invisible to the OS (though that is generally only enterprise-level hardware), I doubt very strongly that there is any way to do what you ask besides interfacing with the specific RAID hardware/software you're using.

Also, while that scheme doesn't greatly increase the risk of lost data, it would almost certainly increase the window that a file is vulnerable - don't forget both drives can write simultaneously and drives are very likely optimized for typical use (meaning sequetial reads and writes). Writing and reading every other sector/cluster/etc could completely invalidate all the asumptions the caching system makes.

Share this post


Link to post
Share on other sites
Sign in to follow this  

  • Advertisement