An example of Veeam Scale-Out Backup Repository mixing whitebox servers

0 Flares Twitter 0 Facebook 0 LinkedIn 0 Email -- 0 Flares ×

Since Veeam announced the Scale-Out Backup Repository technology coming in Veeam Backup & Replication v9, I’ve been asked already multiple times to give some practical examples on how to leverage it. Let’s see together one interesting way to leverage the “performance” policy available in Scale-Out Backup Repository.

A quick look at the available policies

In Scale-Out Backup Repository two policies are available.

In “Data Locality” each backup chain is kept in the same repository, and as soon as a repository is full, a new one is used. The backup admin can add additional repositories and since each chain is managed with one single repository, the machines used in this policy can all be the same. Say you are using a JBOD, you can simply buy another one and add it to the Scale-Out configuration.

Things become more interesting when the policy involved is “Performance”. With it you need at least two repositories, and you can enable each of them to received full backups, incremental backups, or both. Since the size and the IO patterns of a full backup are completely different compared to an incremental, we can plan about using specialised machines based on which backup type they are going to receive.

Let’s see an example: a company has 400 VMs consuming 40 TB of production storage, and creating 6% of daily change rate. The desired retention is 7 days and they will use the Forever Forward Incremental mode. Additional retention will be obtained using Backup Copy jobs sending data to a secondary and even cheaper system (by the way, also backup copy jobs can use a Scale-out Backup Repository).

First, we need some numbers. Using the always great Restore Point Simulator created by my colleague Timothy Dewin we input the desired configuration and we obtain these results:

Rps 01

Design your specialised machines

Before v9 and Scale-Out Backup Repository, we should have designed a single storage machine with a size of 39 TB at least to hold the entire backup data set. This machine should have been able to be at the same time fast enough to ingest incrementals, and large enough to hold all data. With the new Scale-Out solution, we can design a storage group that splits data between incrementals and fulls and use an optimised machine for each role.

In this simulation, incrementals will be received by a super-fast SSD-only machine, high on performance but aso with a high price per GB. But thanks to Scale-Out Backup Repository, we will only need a machine with 7,02 TB of space for the incrementals. Afterwards, incrementals will be injected into a second machine during the transfom operation of the Forever Forward Incremental job. This second storage will be built using large and cheap spinning disks, large enough to hold an entire full and additional space to recreate a second full: the size we need will be 31,89 TB.

Note: I created my configurations using Thinkmate, a US website where you can buy custom-configured servers online. I have no interest in promoting them nor any agreement with them, simply it’s one of the few online resellers offering a free configurator tool. Calculations have been done in September 2015, prices may be different if you read this post in the future.

The “Landing Machine” to ingest incrementals at high speed is going to be a 2U server with 24 * 2,5” disk slots.

2U server

These are the specs:

2 x Eight-Core Intel® Xeon® Processor E5-2640 v2 2.00GHz 20MB Cache
Intel® C602 Chipset – Dual Intel® 10-Gigabit Ethernet (RJ45) – Dedicated IPMI 2.0 LAN Port
6g GB (8 x 8GB) PC3-14900 1866MHz DDR3 ECC Registered DIMM
Boot Drive: 2 x 128GB Micron M600 2.5” SATA 6.0Gb/s Solid State Drive in RAID 1
Storage Drive: 24 x 400GB Toshiba PX03SN Series 2.5” SAS 12.0Gb/s Solid State Drive (MLC)
Controller Card: LSI MegaRAID 9361-8i SAS 12Gb/s PCIe 3.0 8-Port Controller with 1GB Cache + CacheVault Flash Cache Protection Module

Using a RAID 60 configuration on this machine, we can sustain up to 4 concurrent disk failures, have fast rebuild times, yet having 8 TB of storage space to receive backups. With a price of $27,944.00, it means 3,41 USD per effective GB. Not too far from the price per GB of an enterprise array.

For the “Big Machine”, I’ve chosen a 3U machine able to hold 16 large 3.5” disks. For this machine I’ll use 4TB hard drives, they are way cheaper these days than 6TB or 8TB disks and offer shorter rebuild times.

4U server

These are the specs:

2 x Six-Core Intel® Xeon® Processor E5-2620 v2 2.10GHz 15MB Cache (80W)
Intel® C602 Chipset – Dual Intel® 10-Gigabit Ethernet (RJ45) – Dedicated IPMI 2.0 LAN Port
64 GB (8 x 8GB) PC3-14900 1866MHz DDR3 ECC Registered DIMM
Boot Drive:2 x 128GB Micron M600 2.5” SATA 6.0Gb/s Solid State Drive
Storage Drive: 16 x 4.0TB SAS 2.0 6.0Gb/s 7200RPM – 3.5″ – Hitachi Ultrastar™ 7K4000
Controller Card: LSI MegaRAID 9361-8i SAS 12Gb/s PCIe 3.0 8-Port Controller with 1GB Cache + CacheVault Flash Cache Protection Module

The 16 disks are configured again in RAID 60. In this configuration, the machine has an effective storage space of 48 TB and can sustain up to 4 disks failures. With a price of $9,373.00, it means 0,19 USD per usable GB, that is 18 times cheaper than the SSD machine built before. Using maybe Windows 2012 R2 and its deduplication capabilities in addition, we can think that this machine will offer even greater savings.

Final notes

The total price of this “Do it yourself” solution is going to be $37317. Less than 40k to store backups of 400 VM for a week. I’m NOT implying that everyone should start using custom servers for backups and drop any other solution like a dedupe appliance (even if I have some opinions I already expressed in the past), but this is one of the new possible designs that will be available once Veeam v9 and Scale-out Backup Repositories will be available.

Once you have carefully analysed your environment and find out the size of backups you are going to deal with, you will be able to create many different designs using Scale-Out Backup Repository. The design possibilities are endless, which one will be yours?

2 thoughts on “An example of Veeam Scale-Out Backup Repository mixing whitebox servers

  1. Luca, can you also provide a “Fast Machine” calculation with 600GB 6G SAS 10K? This is what I would go with I think, as they are just over $100 these days. SSD still seems like an overkill for a backup repository 😉

    • The same machine with 24 SAS drives is 11743 USD. But the performance between 15k and 7.2k disks in not as much as this design, and in my mind this design is for example a better fit for those environments where the source/production storage is also an all-flash or hybrid, which is becoming more and more common these days.

Comments are closed.