VSAN Backup and Recovery Q&A
I'm just beginning to learn VSAN basics and of course my backup background immediately kicks in. I intend for this to be a living post with additional questions and answers being added in. *To clarify, this information is based on using the VDDK - HotAdd NDB portion of VADP
1) Q: Does VSAN support direct storage backup using VADP bypassing the VM Host running the data (such as how storage can be directly connected to with traditional storage)?:
A: Currently, No, since VSAN does not operate like a traditional storage array using common protocols (iSCSI, NFS or FCoE) direct connection to the storage is not possible - this answer was provided by Jase McCarty via vBrownBag Session @jasemccarty
2) Q: In a multinode VSAN environment, does the VMDKs transfer/copy to the host running the guest VM?
A: No, the VMDK's do not tranfer/follow the guest from one host to another, BUT some data would be cached on the host running the Guest. This probably would not be a full set of info, but primarily the hot blocks from the machine - Per discussion with Edward Haletky@texiwill
3a) Q: Does VSAN communication operate via VMkernel interface?
A: Yes, you need to ensure the type of VMkernel interface is VSAN. - Comment from Edward Haletky@texiwill
3b) Q: If 3a is yes, does it require a dedicated VMkernel interface or does it share the interface (both recommended and actual answers would be nice)?
A: While not required, Backups and VSAN SHOULD be separated onto different NICs to prevent issues with VSAN communication/latency during the high utilization occurring during the backups. - Comment from Edward Haletky@texiwill
4) Q: Assuming a large multinode VSAN deployment where the VMDK files are on a node that is separate from the host running the VM, using VADP would there be any way to control data/communication so as to prevent excessive overhead/bandwidth saturation?
A: Certain VSAN aware backup solutions, such as Veeam, utilizing HotAdd Proxy systems are aware of VMDK locations within the VSAN environment. They will then use a local Proxy system to mount the VMDKs and perform any dedupe/compression locally before transporting to the final destination - Note: This is not available on all backup solutions - Thanks Tim Smith @tsmith_co
5) Q: NetBackup (as an example) can set resource limits based on hosts, datastores, Clusters as well as other determining factors. Since VSAN uses a unified datastore that spans multiple hosts, will VADP be extended to allow backup solutions to better realize data locations in order to improve throughput?
A:
6) Q: When a host is running a guest VM with multiple copies in the VSAN environment, is only one copy active or can the data be read from multiple nodes at the same time? Is this true for both running the guest as a VM as well as using APIs such as VADP for backup and recovery operations?
A:
I'll begin asking around to try and find some answers to these questions and will update the blog accordingly. This could get interesting for my little corner of the VMware/Backup overlap world.
If you have any other questions/answers you would like me to include please send me a message via Twitter!
Thanks goes out to Edward Haletky @Texiwill, Tim Smith@tsmith_co and Jase McCarty @jasemccarty for the answers they provided.
Charles
@whitehattechs
1) Q: Does VSAN support direct storage backup using VADP bypassing the VM Host running the data (such as how storage can be directly connected to with traditional storage)?:
A: Currently, No, since VSAN does not operate like a traditional storage array using common protocols (iSCSI, NFS or FCoE) direct connection to the storage is not possible - this answer was provided by Jase McCarty via vBrownBag Session @jasemccarty
2) Q: In a multinode VSAN environment, does the VMDKs transfer/copy to the host running the guest VM?
A: No, the VMDK's do not tranfer/follow the guest from one host to another, BUT some data would be cached on the host running the Guest. This probably would not be a full set of info, but primarily the hot blocks from the machine - Per discussion with Edward Haletky@texiwill
3a) Q: Does VSAN communication operate via VMkernel interface?
A: Yes, you need to ensure the type of VMkernel interface is VSAN. - Comment from Edward Haletky@texiwill
3b) Q: If 3a is yes, does it require a dedicated VMkernel interface or does it share the interface (both recommended and actual answers would be nice)?
A: While not required, Backups and VSAN SHOULD be separated onto different NICs to prevent issues with VSAN communication/latency during the high utilization occurring during the backups. - Comment from Edward Haletky@texiwill
4) Q: Assuming a large multinode VSAN deployment where the VMDK files are on a node that is separate from the host running the VM, using VADP would there be any way to control data/communication so as to prevent excessive overhead/bandwidth saturation?
A: Certain VSAN aware backup solutions, such as Veeam, utilizing HotAdd Proxy systems are aware of VMDK locations within the VSAN environment. They will then use a local Proxy system to mount the VMDKs and perform any dedupe/compression locally before transporting to the final destination - Note: This is not available on all backup solutions - Thanks Tim Smith @tsmith_co
5) Q: NetBackup (as an example) can set resource limits based on hosts, datastores, Clusters as well as other determining factors. Since VSAN uses a unified datastore that spans multiple hosts, will VADP be extended to allow backup solutions to better realize data locations in order to improve throughput?
A:
6) Q: When a host is running a guest VM with multiple copies in the VSAN environment, is only one copy active or can the data be read from multiple nodes at the same time? Is this true for both running the guest as a VM as well as using APIs such as VADP for backup and recovery operations?
A:
I'll begin asking around to try and find some answers to these questions and will update the blog accordingly. This could get interesting for my little corner of the VMware/Backup overlap world.
If you have any other questions/answers you would like me to include please send me a message via Twitter!
Thanks goes out to Edward Haletky @Texiwill, Tim Smith
Charles
@whitehattechs
3a) Yes, you need to ensure the type of vmkernel interface is VSAN.
ReplyDelete3b) Yes, it is a separate vmkernel interface within ESXi. However, I feel you are asking if it requires a separate physical NIC. In that case, the recommendation is a 10G network link for VSAN. Usually, that implies a dedicated physical NIC.
Note that CBT, snapshot based VADP, and VM AIO (all part of VDDK) will also work with VMs on VSAN. However, if you want to know if you can use a direct mount/access of the VSAN datastore to a backup server, the answer is no. The backup tool needs to use the appropriate API.
In the case of snapshot based VADP or AIO, backups/data collection is performed where the VM is running as they interact direct with the VM itself, such as creating snapshots and mounting those snapshots then getting the data (in the case of VDDK), or in the case of AIO using the SCSI IO Filters.
Check out http://pubs.vmware.com/vsphere-65/topic/com.vmware.sdk.doc/GUID-E2E3BB5D-EE57-4DC5-A5F0-C990A05C6F6C.html
If using Veeam for backups, then it's aware of vSAN. So, when you deploy virtual proxies for "hot add" mode (Virtual appliance mode) for data transport, Veeam is smart enough to use a proxy on the same node as a VMDK file resides in vSAN. The disks are then hot-added into the proxy, which compresses and dedupes the data before sending out the network interface of the proxy VM to the backup repository.
ReplyDelete