cancel
Showing results for 
Search instead for 
Did you mean: 

Partition Failover in EV8?

iternity
Level 4
Partner

Hello,

I have a question to clustering. The setup is an EV 8 server pointing  to two storage devices. When the primary storage device fails, the second one should be used to continue writing data (without any service restart etc).

1.) I was wondering if I can simply set up two partitions, one as open and one as ready. When the first one fails it would go into ready state and the second one would go in the open state. Is something like this possible?

 

2.) Somebody told me about using Building Blocks, but from reading the Admin Guide this seems only a solution for when the EV server fails. We rather want to detect a storage failure and write to a different storage device in that case.

 

3.) Would it be possible to set up MS Clustert that way - e.g. it would detect that a storage device fails rather than EV server and then fail over the UNC Path? If so would this be possbile withouit any service restarts?

 

Thanks a lot for any hints in this direction,

 

bjorn

1 ACCEPTED SOLUTION

Accepted Solutions

JesusWept3
Level 6
Partner Accredited Certified

What you're looking to do *may* be possible but not advisable
However partition roll over is based on Date or Size restrictions

So for instance on the date one, some people like to have partitions based on a particular year or quarter, so you could have 2011 Q1 that rolls over in march to 2011 Q2 etc

However there is no roll over based on whether the storage is available or not.

Another thing, is the storage going to be mirrored so the same data is available on both storage devices? because if it is, then the idea of the partition roll over isn't exactly necessary, but if it isn't then you would just be writing new data and your old data wouldn't be accessible

The problem with the approach at that point would be that you have fragmented datasets that are out of sync, so what then happens if the primary storage becomes available again? all the data written to the fail-over storage would have to come across and is only really good if the storage is the exact same path.

Last but not least if you were going to try and test this and in the event of actual failovers also, you would end up with many small partitions, and also if the storage isn't mirrored or replicated then you would lose some form of SIS along the way, and having mentioned SIS, i'm not quite sure how EV would react if it believes that an attachment has already been stored and what will happen when it tries to verify it.

 

For instance if you have E:\Vault Stores\Ptn1 and you have a document shared called myWordDoc.docx thats 1.2MB and then you fail over to E:\Vault Stores\Ptn2 and the same document comes in, EV would normally say this item is the same as an item already stored, so i'm going to share it

I don't know whether it would error because the item is not available, or whether it would simply just say "the items there, i'm not going to check whether its available" and then when you try to open the item again it errors out because the sis part isn't available.

Either that or it may check for the existance and then it says not available so i'm going to share this as a new file, and now you have duplicated data in your storage

 

As for the cluster option, i believe you can have HA (High Availability) set for storage failures, however part of the cluster failover will be to stop services on one server and then start on the other server (if it can do it gracefully enough)


To be honest though, what you want to do with clustering and storage isn't advisable but what i would suggest is having a look in to the VMWare ESX fail over solutions that are a lot quicker and a lot more seamless and look at having replicated or mirrored storage, though i believe the cost for those would go up some.

Personally i would invest more in having the storage be more highly available and concentrate less on the EV servers

https://www.linkedin.com/in/alex-allen-turl-07370146

View solution in original post

6 REPLIES 6

FreKac2
Level 6
Partner Accredited Certified

AFAIK no, it's partition rollover not failover :)

So it's triggered by a date or amount of free space, not offline/online states.

I think you would be better off and let the storage device, storage processors and/or storage software deal with this kind of failure.

Maverik
Level 6

I wonder if you would actually be better off with some sort of SAN solution that has redundency built in..... I am no SAN expert I am afraid though, but would think this was the best method.

As far as partition rollover is concerned this is not really what is was designed for.  Also if the old storage went offline and the new one came onlien writing new data to the new location surely then the user's would not be able to access the old data for the storage that is offline???? There is no way to automate the rollover on storage failure.

Building blocks is for the EV servers as you correctly say.

MS cluster I guess would be possible if you were pointing to a share in the cluster, but again you are not protecting the core data and would still need some SAN resiliance.  Yes Server A faills over and Server B take it on but they would be sharing the same storage no.... This may explain more.

http://technet.microsoft.com/en-us/library/cc785197(WS.10).aspx

JesusWept3
Level 6
Partner Accredited Certified

What you're looking to do *may* be possible but not advisable
However partition roll over is based on Date or Size restrictions

So for instance on the date one, some people like to have partitions based on a particular year or quarter, so you could have 2011 Q1 that rolls over in march to 2011 Q2 etc

However there is no roll over based on whether the storage is available or not.

Another thing, is the storage going to be mirrored so the same data is available on both storage devices? because if it is, then the idea of the partition roll over isn't exactly necessary, but if it isn't then you would just be writing new data and your old data wouldn't be accessible

The problem with the approach at that point would be that you have fragmented datasets that are out of sync, so what then happens if the primary storage becomes available again? all the data written to the fail-over storage would have to come across and is only really good if the storage is the exact same path.

Last but not least if you were going to try and test this and in the event of actual failovers also, you would end up with many small partitions, and also if the storage isn't mirrored or replicated then you would lose some form of SIS along the way, and having mentioned SIS, i'm not quite sure how EV would react if it believes that an attachment has already been stored and what will happen when it tries to verify it.

 

For instance if you have E:\Vault Stores\Ptn1 and you have a document shared called myWordDoc.docx thats 1.2MB and then you fail over to E:\Vault Stores\Ptn2 and the same document comes in, EV would normally say this item is the same as an item already stored, so i'm going to share it

I don't know whether it would error because the item is not available, or whether it would simply just say "the items there, i'm not going to check whether its available" and then when you try to open the item again it errors out because the sis part isn't available.

Either that or it may check for the existance and then it says not available so i'm going to share this as a new file, and now you have duplicated data in your storage

 

As for the cluster option, i believe you can have HA (High Availability) set for storage failures, however part of the cluster failover will be to stop services on one server and then start on the other server (if it can do it gracefully enough)


To be honest though, what you want to do with clustering and storage isn't advisable but what i would suggest is having a look in to the VMWare ESX fail over solutions that are a lot quicker and a lot more seamless and look at having replicated or mirrored storage, though i believe the cost for those would go up some.

Personally i would invest more in having the storage be more highly available and concentrate less on the EV servers

https://www.linkedin.com/in/alex-allen-turl-07370146

iternity
Level 4
Partner

Hello all,

 

thanks for your answers, that has been very insightful.

 

bjorn

JesusWept3
Level 6
Partner Accredited Certified
don't forget to select something as a solution if you feel its been helpful :)
https://www.linkedin.com/in/alex-allen-turl-07370146

ggeorgi
Level 5
Partner Accredited

You need to implement a storage mirroring solution.... this means that disks from different storages, will belong in a mirror set so if the disk of the first storage or the whole storage, fails it will continue writing or reading from the second disk. I do not want to mention any brands in this forum. Maybe just google the "storage mirroring software". Unfortunately it is not cheap solutions.