Skip to main content

Going Backup-less with Exchange 2010?

 Going Backup-less with Exchange 2010?



Microsoft Exchange Server 2010 has attracted a lot of attention because of its new database availability group (DAG) feature. DAGs offer a new spin on the standard Exchange database model by letting you maintain multiple, continuously updated copies of mailbox databases on multiple servers without requiring shared storage or the use of SANs.

This new feature isn't without controversy, though, because it offers the possibility of creating a system that's designed to run without making frequent backups for database restoration or recovery. The notion of routine operations without backups has made a lot of people nervous, so I wanted to talk this week about whether backup-less operation is really possible, not to mention safe.

The basic idea is simple: If you maintain enough copies of a particular mailbox database, you don't need to make frequent backups because you'll always have an available copy. The magic number in this case is 3: 3 copies of each protected database is what Microsoft claims to be sufficient. With only two copies, you'd still be vulnerable to the loss of a single machine, but three independent copies, on three separate physical machines, gives you the ability to withstand two simultaneous failures, which seems like it should be enough.

There are two issues that make using DAGs instead of routine backups a challenging proposition to accept. The first problem is cost. Maintaining three copies of a database necessarily implies having three servers to put that copy on. Running three servers means three licenses of Exchange 2010, plus three licenses of Windows Server 2008 R2 Enterprise edition or Server 2008 SP2 Enterprise edition—and Enterprise is much more expensive than the Standard edition. However, Windows failover clustering is available only in the Enterprise edition, and the DAG feature depends on it. Then there's the hardware cost, which admittedly might be cut down by intelligent use of virtualization—keeping in mind that putting all your DAG copies on separate VMs in the same physical machine or data center takes away much of the benefit of using DAGs in the first place!

The second problem is a little more complicated. Exchange 2010, like its predecessors, creates transaction logs that contain records of every transaction applied to a given database. Putting a mailbox database into a DAG doesn't change that, which means that logs will still continue to accumulate until you do a full backup of the database. For that reason, Microsoft recommends that you enable circular logging on those databases. The very term circular logging makes many experienced Exchange administrators nervous because they know that without logs, your database recovery options are limited and painful. That lack of logs seems to be a bigger sticking point for many customers than the additional cost of DAG-based deployments. However, the DAG mechanism itself ensures that the logs are kept until the transactions therein have been committed on all remote copies.

What I've found a few sites doing is taking a hybrid approach: deploying DAGs but leaving circular logging turned off, and doing regular full backups, but on a less frequent schedule. This method offers the comfort of regular backups without as much overhead, while at the same time preserving the utility of DAGs. You can change the frequency of your backups as much as you like to find the right balance. Then when you're comfortable with your DAG implementation (and, most importantly, with how you restore data when necessary), flip the circular logging switch for your DAG databases and cut back the backup frequency yet again.
I like this approach, and it's one that I'll be recommending, but I'm curious: What do you think about the possibilities of going without routine backups? Does it make you nervous? Drop me a line to let me know.

Comments

Popular posts from this blog

Integration with vCloud Director failing after NSXT upgrade to 4.1.2.0 certificate expired

  Issue Clarification: after upgrade from 3.1.3 to 4.1.2.0 observed certificate to be expired related to various internal services.   Issue Verification: after Upgrade from 3.1.3 to 4.1.2.0 observed certificate to be expired related to various internal services.   Root Cause Identification: >>we confirmed the issue to be related to the below KB NSX alarms indicating certificates have expired or are expiring (94898)   Root Cause Justification:   There are two main factors that can contribute to this behaviour: NSX Managers have many certificates for internal services. In version NSX 3.2.1, Cluster Boot Manager (CBM) service certificates were incorrectly given a validity period of 825 days instead of 100 years. This was corrected to 100 years in NSX 3.2.3. However any environment originally installed on NSX 3.2.1 will have the internal CBM Corfu certs expire after 825 regardless of upgrade to the fixed version or not. On NSX-T 3.2.x interna...

Calculate how much data can be transferred in 24 hours based on link speed in data center

  In case you are planning for migration via DIA or IPVPN link and as example you have 200Mb stable speed so you could calculate using the below formula. (( 200Mb /8)x60x60x24) /1024/1024 = 2TB /per day In case you have different speed you could replace the 200Mb by any rate to calculate as example below. (( 5 00Mb /8)x60x60x24) /1024/1024 =  5.15TB  /per day So approximate each 100Mb would allow around 1TB per day.

Device expanded/shrank messages are reported in the VMkernel log for VMFS-5

    Symptoms A VMFS-5 datastore is no longer visible in vSphere 5 datastores view. A VMFS-5 datastore is no longer mounted in the vSphere 5 datastores view. In the  /var/log/vmkernel.log  file, you see an entry similar to: .. cpu1:44722)WARNING: LVM: 2884: [naa.6006048c7bc7febbf4db26ae0c3263cb:1] Device shrank (actual size 18424453 blocks, stored size 18424507 blocks) A VMFS-5 datastore is mounted in the vSphere 5 datastores view, but in the  /var/log/vmkernel.log  file you see an entry similar to: .. cpu0:44828)LVM: 2891: [naa.6006048c7bc7febbf4db26ae0c3263cb:1] Device expanded (actual size 18424506 blocks, stored size 18422953 blocks)   Purpose This article provides steps to correct the VMFS-5 partition table entry using  partedUtil . For more information see  Using the partedUtil command line utility on ESX and ESXi (1036609) .   Cause The device size discrepancy is caused by an incorrect ending sector for the VMFS-5 partition on the ...