Is writeback mode stable or not about enhanceio

[onlyjob]

I only used EnhanceIO for read caching (safe).
For write caching I would use dm-writeboost which is more robust.

[samuelxhu]

@onlyjob Very interesting information. Just wonder whether or not you still use EnhanceIO for read cache, because AFAIK dm-writeboot can be configured as combined read-and-write cache as well.

Should i focus on dm-writeboost only, or still there is a good reason for EnhanceIO as read cache?

[onlyjob]

I don't use EnhanceIO any more. I found read performance acceptable once write stress is relieved by dm-writeboost or by ext4 external journal. Also in context of distributed parallel storage read caching is not very effective due to highly randomised access...

[samuelxhu]

@onlyjob I just looked around by googling but not able to find much information on dm-writeboost. Thus i turn to you for further clarification.

1. In your view, is dm-writeboost stable enough for production? I see it use rambuf to aggregate multiple writes before flushing to SSD, so it relies much on the correctness of barrier/flush commands from the upper software layer, otherwise there may be data loss, right?
2. Do you use dm-writeboost in production your self? for howlong? It says quite a lot when there are actual use cases.

I have a production of two Gluster nodes with HDDs, now I plan to add SSDs to boost the overall performance.

[onlyjob]

1. Yes, I believe dm-writeboost is mature enough for production.
   I think author of dm-writeboost is better qualified to answer design questions than me.

2. It trust dm-writeboost and I use it in production. I could not build enough confidence in (almost) unmaintained EnhanceIO to rely on it. I believe dm-writeboost is more robust and it is well maintained. I've been using dm-writeboost ever since its first release. There were only few issues in earlier versions that appears to be all fixed.

I use dm-writeboost with RAM disk as caching devices.
As for Gluster I do not trust it due to lack of integrity checks...

[lanconnected]

Hi!
We are using enhanceio in writeback mode for more than a year in production. There were some stability issues hopefully resolved now, and we've never lost any data even after power losses/hard resets. We will support this branch and try to address any reported problem, but only for RHEL6/RHEL7 and OpenVZ kernels for now. dm-writeboost vs enhanceio discussion is probably out of topic here, they both have their pros and cons. If you use and enhanceio compiled from this fork and have any problem, please file the issue here, any feedback is welcome.

[samuelxhu]

@lanconnected This is great. Thanks a lot.