Poor man's / Opensource - High availability storage? maybe zfs

I

intel

n00b
Joined
Apr 5, 2009
Messages
35
I am curious if anybody knows if its possible to implement a cheap/affordable high availability storage solution?

The perfect system, in order to be reliable and highly available should be able to:
  • High performance with multiple drives, maybe SSD L2 cache like ZFS (would love to implement ZFS let me know if you know how)
  • IP sharing with other "sister" box or somesort of automatic failover at NAS level (been reading about FreeBSD CARP interesting stuff)
  • Automatic Promotion, re-sync of data once master node is back online - Possible "maintenance mode" to upgrade ram then resync data then promote to master again?
  • IP Multi-path maybe?

I think those are the main points. The idea behind this is that I have a vmware ESXi datastore in a single local attached drive. Performance is poor. We are looking to move more machines to ESXi and save on electricity $$$

I understand there is more than a single point of failure even if storage is rock solid (vMotion could come into play here) -- however, I would like to focus on storage for this topic.

Some scenarios that I play in my head that could cause downtime and I would like to avoid:

  • Data Loss (Critical) system needs to be able to keep consistency with at least 1 failed drive
  • Scheduled System Hardware upgrades - Minimal impact at the time to upgrade ram. Shutdown master server, slave takes over immediately, if they share same IP downtime should be minimal to ESXi (maybe a better way?)
  • Complete failure (Unscheduled downtime) - sometime a motherboard goes bad, memory crashes server. Slave should take over immediately
  • Data inconsistency between slave/master "Split brain" -- Either master or slave should know who was the last one to have the latest copy and replicate it back.

I know that in order to install an OS remotely to an ESXi box, the ESXi needs to have a remote iSCSI data store. I have a Dell PowerConnect 3548 at my disposal link to Dell

I am pretty sure I am not alone in this scenario. We all want to minimize data loss, downtime and the like. Please share if you have any ideas or know of any way to achieve this on a very very tight budget. Storage size is not of concern right now, maybe 3TB worth of actual ESXi image disk size.

Thanks
 
ZFS via SE11/napp-it would do everything you want, except HA/Failover (at least not in an easy to manage way). The only way I know of this being implemented is via (non-free) NexentaStor, with and extra plugin for HA/Failover - one for shared storage, and one for replicated storage. Obviously not fitting into the "poor man's" requirement.
 
Astronot said:
ZFS via SE11/napp-it would do everything you want, except HA/Failover (at least not in an easy to manage way). The only way I know of this being implemented is via (non-free) NexentaStor, with and extra plugin for HA/Failover - one for shared storage, and one for replicated storage. Obviously not fitting into the "poor man's" requirement.
Click to expand...


Wow $5,000 + for the Nexenta license to do replication! thanks for chipping in
 
Ive been looking for something similar. Iv yet to find a good solution for zfs. Found one with linux though.
 
Don't stand up yet; Available storage includes the raw drive capacity of all drives in your pool. While this doesn't count l2arc/ZIL drives, it does count any parity or mirror devices.

See this.

A machine I recently put together has 30 3TB drives. Effectively ~60TB, by their view, 90TB, which is about $15000 just for a silver license with no plugins.

On a lighter note, you can get a mostly functional CE edition for up to 18TB for free. Considering the above calculation, it's basically an 8-drive 2TB array or a 6-drive 3TB array - any more and you must pay.
 
I've thought of trying this, but not sure how well it would work or if it's even possible, but maybe test it out and report back:


Have 3 storage servers setup with equal number/size of arrays, export as iSCSI volumes. Each server would map the other two's iSCSI luns. With all the luns in place, raid 5's could be done accross all 3 servers. Only one machine at a time would be able to run the raid, but if one goes down you should be able to restart the raid on one of the other two, and it would be a degraded raid 5.

No idea how or if this would work, though. I'd actually be curious to see it in action. They could be behind a NAT and then you forward NFS or w/e port to the box acting as master and if this changes you just change the forward. Perhaps this could be automated. This way your vmware box is always pointing to same IP. Could probably do something with DNS too.
 
I hate to say it, but poor mans NAS Storage and automatic fail over do not belong in the same sentence. 5000$ may sound like alot, but most entry level enterprise SANS cost 20K and up. You may want to look into Nexenta if you think your storage pool will not grow beyond 18TB of RAW data. Its a pretty decent product.
 
Vengance_01 said:
I hate to say it, but poor mans NAS Storage and automatic fail over do not belong in the same sentence. 5000$ may sound like alot, but most entry level enterprise SANS cost 20K and up. You may want to look into Nexenta if you think your storage pool will not grow beyond 18TB of RAW data. Its a pretty decent product.
Click to expand...

HA is a comercial extra-plugin, only available for NexentaStor Enterprise (not for free 18TB Community edition)
see http://nexenta.com/corp/component/virtuemart/?page=shop.browse&category_id=7
http://nexenta.com/corp/nexentastor-overview/nexentastor-versions

Gea
 
Last edited:
You must log in or register to reply here.
Back
Top