Veritas dirty region logging
#1
Scooby Regular
Thread Starter
Join Date: Mar 2001
Location: Berkshire
Posts: 5,528
Likes: 0
Received 0 Likes
on
0 Posts
Does anyone have any documents on DRL (sequential or random) when used with a Veritas filesystem ??
I know basically how to set it up and that it is useful for mirror reconstruct times as it keeps a log so that if a sys crashes when data is being written to the mirror you dont have to resync the whole mirror, only what is inconsistent.
Cheers
Dave
I know basically how to set it up and that it is useful for mirror reconstruct times as it keeps a log so that if a sys crashes when data is being written to the mirror you dont have to resync the whole mirror, only what is inconsistent.
Cheers
Dave
#2
What did you want to know ? I've included a brain dump below, if you want any other information/clarification let me know and I will dig out my Advanced Veritas VxVM/VxFS course notes.
Michael
Dirty Region Logging (DRL) subdisks are very small -- 2 to 10 disk blocks (i.e. 1KB to 5KB).
DRL purpose in life is to limit the amount of work the VxVm has to do when resyncing a mirrored volume after a system crash. Without DRL, VxVM would have to resilver the entire mirror to the other. With DRL, it syncs only those regions marked as "dirty" in the Dirty Region Log. This can significantly speed up sync time after recovering from a system crash.
The downside is that it may slow system performance during normal operation, as the DRL is constantly updated (written to) by VxVM. DRL's are bitmaps eith a bit mapped to a "chunk" of space in the mirrored volume the DRL is attached to.
Say you have an 800MB volume. And you create a 2 block (1KB) log subdisk. There are 8000 bits in one KB, so each bit of the DRL will map to 1MB of the volume, the larger the DRL that is defined, the finer grained the DRL map becomes. At initialization, all bits in the DRL are set to Zero. Every time the mirrored volume is written to the corresponding bit in the DRL which maps to the region being written to marked as dirty by being set to 1. Periodically, the VxVM rezeros bits as appropriate.
So in the event of a system crash, VxVm only has to resilver that portion of the mirrored volume which has the corresponding DRL bits set.
A separate DRL (log plex) has to be specified for each volume. The DRL (log plex) can be added or removed from the volume at any point in time.
To add a log plex, use the command:
#vxassist -g <dg> addlog <vol>
To remove a log plex, use the command:
#vxplex -g <dg> -o rm dis <plex>
You may have more than one DRL per mirrored volume, in order to "mirror" the DRL, although the benefits are marginal. (If the last DRL fails, then Dirty Region
Logging is disabled, and this only impacts resync times in the unlikely event that the system crashes before this is resolved).
Michael
Dirty Region Logging (DRL) subdisks are very small -- 2 to 10 disk blocks (i.e. 1KB to 5KB).
DRL purpose in life is to limit the amount of work the VxVm has to do when resyncing a mirrored volume after a system crash. Without DRL, VxVM would have to resilver the entire mirror to the other. With DRL, it syncs only those regions marked as "dirty" in the Dirty Region Log. This can significantly speed up sync time after recovering from a system crash.
The downside is that it may slow system performance during normal operation, as the DRL is constantly updated (written to) by VxVM. DRL's are bitmaps eith a bit mapped to a "chunk" of space in the mirrored volume the DRL is attached to.
Say you have an 800MB volume. And you create a 2 block (1KB) log subdisk. There are 8000 bits in one KB, so each bit of the DRL will map to 1MB of the volume, the larger the DRL that is defined, the finer grained the DRL map becomes. At initialization, all bits in the DRL are set to Zero. Every time the mirrored volume is written to the corresponding bit in the DRL which maps to the region being written to marked as dirty by being set to 1. Periodically, the VxVM rezeros bits as appropriate.
So in the event of a system crash, VxVm only has to resilver that portion of the mirrored volume which has the corresponding DRL bits set.
A separate DRL (log plex) has to be specified for each volume. The DRL (log plex) can be added or removed from the volume at any point in time.
To add a log plex, use the command:
#vxassist -g <dg> addlog <vol>
To remove a log plex, use the command:
#vxplex -g <dg> -o rm dis <plex>
You may have more than one DRL per mirrored volume, in order to "mirror" the DRL, although the benefits are marginal. (If the last DRL fails, then Dirty Region
Logging is disabled, and this only impacts resync times in the unlikely event that the system crashes before this is resolved).
#4
Scooby Regular
Join Date: Oct 2000
Location: Surrey, UK
Posts: 8,384
Likes: 0
Received 0 Likes
on
0 Posts
check out http://www.cuddletech.com/veritas/
theres some good stuff on there...
[Edited by DrEvil - 9/10/2002 12:16:03 PM]
theres some good stuff on there...
[Edited by DrEvil - 9/10/2002 12:16:03 PM]
Thread
Thread Starter
Forum
Replies
Last Post
ossett2k2
Engine Management and ECU Remapping
15
23 September 2015 09:11 AM
Welloilbeefhooked
Engine Management and ECU Remapping
5
12 September 2015 05:32 PM