Direct SAN Access vs Virtual Appliance

[ejleipold]

Hi Guys,

I'm in the process of replacing/refreshing my vSphere HA Cluster (5.0) with all new equipment to go to 5.5 (including an IBM V3700 SAN).

In my current environment I'm just using the Virtual Appliance transport mode (Veeam server running in a VM) but I was wondering if it was worth going down the Direct SAN access path for my new environment and wanted to know if the performance between direct SAN access and the Virtual App is worth the increased level of complexity?

Thanks

EvanL

[Dude]

Direct SAN means lower CPU load on your ESXi servers, lower production network utilization (because all backup traffic goes through your storage network), and most likely faster backup times. If none of those things will make a valuable positive impact to you then Direct SAN won't make a practical difference in your environment.

[foggy]

Generally, depending on the environment, direct SAN can be faster than hotadd. It's hard to predict as there are too many factors involved in here that affect data transfer speed, so only actual testing can answer this question.

That said, I would recommend to retain virtual proxies for the sake of faster restores through hotadd, since direct SAN cannot be used for restores.

[ejleipold]

Good to know, thanks guys.

Is there any real gotchas when connecting a proxy to the SAN (Besides making sure auto-mount is disabled)?

[veremin]

Nope, as VB&R takes care about it. Thanks.

[Treeeman]

Hello,

i have a EMC FC Storage, 10 ESXi Server i a VMware HA Cluster with about 150 VMs running.
I want to implement Veeam. I plan to use a physical Veeam Backup Server with "Direct SAN Access".

Question:
Are there any Benefits using instead "Virtual Appliance" Transport Mode? Faster? Easier?

Thanks.
Kind Regards
Marco

[Shestakov]

Hello Marco,
In general Direct SAN is the fastest transport mode, while Network mode is the slowest one and not recommended on 1Gb Ethernet.
However in some environments Virtual Appliance mode shows the best results. Please review the topic for more info and ask additional questions if you have any.
Thanks!

[Vitaliy S.]

I would go with direct SAN access just not to load your ESXi hosts with data processing tasks. my 2 cents.

[Treeeman]

Hi,

thanks for your fast reply.
So if i read the upper posts i will use direct SAN for Backup and deploy additionally a virtual appliance proxy.
Like "foggy" wrote this is reccomended to speed up my restores with this virtual appliance proxy.

Thanks!
Kind Regards
Marco

[Shestakov]

Hi Marco,
True, leaving a virtual proxy is a good idea for the restore purposes.
Thanks!

[Treeeman]

Hi!

i thought about this issue a little bit more. Please correct me if i am wrong.
Let's assume there is no performance difference between Direct SAN Access and Virtual Appliance Transport Mode. What are the advantages and disadvantages from one Virtual Appliance (with 2 vCPU) on each ESXi?

advantages:
- No need for SAN Connectivity for my physical Backup Server with attached Tape Library (no SAN Zoning or FC Cards needed)
- I am able to run more simultaneous Jobs because my Virtual Appliance will have total more CPU Cores like my physical Server

disadvantages:
- CPU and RAM Ressources from my ESXi Hosts will be used for Backup
- Backup Traffic over Network because (compressed) Backup Data will be transfered from Virtual Appliance to my physical Backup Server over Network
- I have to monitor and update (e. g. Windows Updates) each Virtual Appliance

Are there some Points i should put on my List?
I am not sure if my second Point from my advantages list is really a advantage. I think this also depends on my Backup Repository. If my 10x Virtual Appliances with 2 vCPU are used at the same time this will mean i will be able to do theoretical 20 tasks. This will mean there is a lot of network and storage traffic. So my backup repository should be able to handle this with enough IOPS. Any ideas how much IOPS my Backup Repository should have?

Thanks for your help.
Kind Regards
Marco

[Vitaliy S.]

- I am able to run more simultaneous Jobs because my Virtual Appliance will have total more CPU Cores like my physical Server

Yes and no. The total number of jobs you can run will depend on the observed bottleneck statistics in both configurations. For example, if bottleneck will be target, then amount of CPU resources of the proxy server will not matter.

- I have to monitor and update (e. g. Windows Updates) each Virtual Appliance

Not sure I understand this disadvantage, since you should do that for all Veeam components.

Are there some Points i should put on my List?
I am not sure if my second Point from my advantages list is really a advantage. I think this also depends on my Backup Repository. If my 10x Virtual Appliances with 2 vCPU are used at the same time this will mean i will be able to do theoretical 20 tasks. This will mean there is a lot of network and storage traffic. So my backup repository should be able to handle this with enough IOPS. Any ideas how much IOPS my Backup Repository should have?

You really need to run jobs in both configs and see where the bottleneck will be. After knowing this fact it will help you to optimize your backup configuration further. The better performance you have for the repository, the better it would be. Please review this topic for more info > Calculate IOPS for Backup Repository

[Treeeman]

Thanks Vitaliy!

[bgagnon]

At my previous employer, I implemented Direct SAN Backup with a dedicated Storage Server (just a box I put together that was about 20TB, had 32GB of RAM, some 8 core processors and some SSD boot disks) and that was probably the highest performing for backups. You can crank up compression to the max and not worry about it affecting your VMware Environment, the only network usage is over the storage network so the actual ethernet / LAN fabric isn't being saturated by backup traffic. I'm going to go through the works of trying to convince my new employer to go that route as well here in the near future.

That said, there is a place for Virtual Appliances and Hot Add, specifically if you don't want to spend the money on a 'dedicated' piece of hardware for backups (however, you're probably and hopefully backing up to different storage anyway).

All in all it really depends on your budget and where you want to go with it. If Veeam is your primary backup solution, I would dedicate a piece of hardware to it and use Direct SAN - that way if your entire environment tanks you still have some method of control with your backups (hopefully that never happens).

Really what's nice about it is that you have options - there are a few ways to go. I did Direct SAN Backup at my old employer with a Virtual Appliance in each cluster for restores (I think I saw that suggested when I skimmed this thread). Veeam was my primary backup at that place and because of that we invested in it, not only the software but also in some decent hardware for it - but it totally depends on your budget and what you're willing to throw at it.

[rborhara]

If you are on Vsphere 5.1 or older then watch out for the Lun limit needing to be less than 2Tb for the backups to work in Virtual Appliance mode

My backups run faster in VA mode than direct SAN not sure why, ISCSI is all setup and working fine, I just dont think vmware knows how to distinguish between 10GB connections when both the management netowrk is 10gb and the san is on a seperate 10GB sAn switch

[Gostev]

My backups run faster in VA mode than direct SAN not sure why, ISCSI is all setup and working fine, I just dont think vmware knows how to distinguish between 10GB connections when both the management netowrk is 10gb and the san is on a seperate 10GB sAn switch

Actually, direct SAN backups are performed directly from the storage (VMware is not involved), so there results would indicate a hardware configuration issue on your direct SAN backup proxy server, or in its connectivity to SAN.

[rborhara]

Veeam actually does use my san infrastructure (2x nics dedicated to iscsi transfer show data transfer rate although one of the san dedicated nics has more throughput than the other most probably due to the equallogic 6510x san support active passive and not active / active