I've read a few posts with similar questions but wanted to confirm I'm doing the right thing regarding our new backup repository. Was hoping someone could give me a few pointers We're using B&R Enterprise Plus v7 patch 3 on vSphere 5.5.
We've recently installed 6 x 2TB drives into a host to be used for our backup repository. The host is a DELL R710 which has a PERC 6/I RAID controller. First question is the recommended RAID level. I've read that for safety and performance RAID 10 is the best option but this takes out half the disk space. They are 2TB NL-SAS drives (so basically SATA) and VMWare recommend formatting local storage using SATA as RAID 6. I believe this is due to read errors that occur more frequently on SATA drives. I'm sure someone in the community has a similar setup so would be curious to see what others are using as their RAID level. We have configured it as one large LUN formatted using RAID 5 but reckon this should be changed now.
Final question. We have our Veeam server running on Server 2012 as a VM. The local storage is setup as a VMFS datastore and we have added a drive to the VM formatted with NTFS. Now I did read that this is not the best option (in another post) for a backup repository and that maybe an RDM directly from the server is the way to go. Not quite sure how to configure this though, unless we use the in guest iSCSI initator to connect to the local RAID array ?
Any help / advice greatly appreciated! We're just testing this now so can blow away the RAID config and start again if need be.
Assuming I understood your scenario correctly, you're going to use one storage as a both production datastore and a backup repository, right? If so, please, be aware it's highly unrecommended to do so, because with such strategy in place you would have all eggs in one basket. In other words, if something happened with the underlying device you would loose both production VMs and their backups.
Thanks for the reply. Nope, we're using this local storage for backups only. The Veeam server runs from an Equallogic SAN along with the rest of our prod data.
Thank you for the clarification. What backup mode you're planning to use? Forward or reversed incremental one? Typically, we recommend using RAID 10, because, Both 5 and 6 has a certain IOP penalty. This might result in decreased backup performance, especially, if reversed incremental mode is used due to its I/O intensive nature. Thanks.
We've been using reverse incremental on our existing server so will stick with that. Thanks for the info, I figured RAID 10 was the way to go, but it's a big trade off in terms of disk space usage. We're only an SMB though and our current backup size is 1.5TB so may stay with RAID 5 for the extra space.
I suggest you running several tests with reversed incremental mode coming to RAID 5 and see whether you're ok with backup performance you get. Even if the resulting performance is far from ideal, but the backup activity doesn't exceed the backup window, you might still keep your deployment as is. Otherwise, the forward incremental might the way to go. Thanks.
leebtish wrote:We have our Veeam server running on Server 2012 as a VM. The local storage is setup as a VMFS datastore and we have added a drive to the VM formatted with NTFS. Now I did read that this is not the best option (in another post) for a backup repository and that maybe an RDM directly from the server is the way to go. Not quite sure how to configure this though, unless we use the in guest iSCSI initator to connect to the local RAID array ?
Neither RDM not virtual disk is recommended to store your backup files, because if you lose the ESXi server that hosts backup server, you will not be able to access/download your backup files. The only way to do that would be to setup another ESXi host and then read data from VMFS datastore.
Thanks for the info Vladimir / Vitality. Will check performance and see how we go. On the RDM / virtual disk front, I see your point. Unfortunately we needed another host for our cluster and went with this option. Would have been nice to have a standard Windows server and an NTFS volume for the repository but didn't have that available. We will be copying backups to tape so should be able to recover from that if we lose the host.
I use an HP server with direct attached (SAS) disk arrays. I have them all setup in RAID5 and the performance is acceptable. These are 2Tb 7200RPM drives and I have over 60Tb of storage space. Performance is much better than using SAN storage which I was doing before (SAN dedicated only for backup data). I think RAID10 would probably give you the best performance but the space sacrifice is too great. We perform backup copies of all our backup data to a secondary location so I am not terribly concerned with the loss of a drive as I can recover so that is why I use RAID5.
We use the options Enable Synthetic Full & Transform previous chains into rollbacks every Friday night. The transforms take quite a while on a few of our backups as there is a pretty good amount of changed data each day (100+Gb/day for a single VM). Those take a while to roll up (12+hours). The others are usually pretty quick.
Something to consider is if you are going to tape out. Your tape drive must be connected to your B&R master server so if you are using physically attached tapes you can't use a VM unless you find a way to iSCSI attach the drives to the VM.
Since my backup server is all in 1 and physical I am fine. My tape drives are FC attached through a switch to my backup server. This runs very fast as well.
Thanks a lot for your post. Good to know someone else is using RAID 5 and performance is acceptable. With the tape drive it will be directly connected to the server and I'm hoping we can configure it using iSCSI. I've read this is possible on the VMWare forums so we'll see how it goes.
I ran a test job and the throughput was 5 x faster than our current backup server so looking good at the moment
I am considering going to local storage as well for some of the same reasons you have. I would like to know what type of dedupe you're seeing using Veeams software dedupe?
I thought I would share some of my Windows 2012 R2 deduplication numbers.
I have 3 drives on a physical HP server. Each drive has a 18.1Tb capacity.
Each has deduplication turned on and set-up to run every 1 days.
I did exclude some of my backups. I exclude backups with a VBK larger than 1 Tb. That would be about 5 out of 25 jobs I run.
All of my jobs are forever incremental with a transform to full on Friday night.
The jobs all use inline deduplication, no compression and WAN Optimized block size.
Drive 1: 11.1Tb Used after Dedup / 17.7 Used Before Dedupe
Drive 2: 6.88Tb Used after Dedup / 11.8 Used Before Dedupe
Drive 3: 11.0Tb Used after Dedup / 14.6 Used Before Dedupe
I did noticed that some of my Friday transforms take longer than expected. I have not confirmed that this is due to re-hydration of the file yet but suspect it may be part of the issue.
Windows deduplication apart you can use 50% as a data reduction ratio (compression + dedupe) that can be achieved with VB&R, because typically you won't see ratios less than that. However, it depends heavily on type of data stored inside VM (for instance, you wouldn't expect impressive data reduction rates from VM storing jpeg images)
nunciate wrote:I use an HP server with direct attached (SAS) disk arrays. I have them all setup in RAID5 and the performance is acceptable. These are 2Tb 7200RPM drives and I have over 60Tb of storage space. Performance is much better than using SAN storage which I was doing before (SAN dedicated only for backup data). I think RAID10 would probably give you the best performance but the space sacrifice is too great. We perform backup copies of all our backup data to a secondary location so I am not terribly concerned with the loss of a drive as I can recover so that is why I use RAID5.
We use the options Enable Synthetic Full & Transform previous chains into rollbacks every Friday night. The transforms take quite a while on a few of our backups as there is a pretty good amount of changed data each day (100+Gb/day for a single VM). Those take a while to roll up (12+hours). The others are usually pretty quick.
Something to consider is if you are going to tape out. Your tape drive must be connected to your B&R master server so if you are using physically attached tapes you can't use a VM unless you find a way to iSCSI attach the drives to the VM.
Since my backup server is all in 1 and physical I am fine. My tape drives are FC attached through a switch to my backup server. This runs very fast as well.
Hi Nunciate,
I've just installed the HP D2600 Storage array on the physical server Win 2012 R2.
How do you set the 60 TB LUNs ?
I'm interested to know about the stripe size, the NTFS allocation unit size and the backup settings that you are using.
--
/* Veeam software enthusiast user & supporter ! */
We're using B&R Enterprise Plus v7 patch 3 on vSphere 5.5.