MacPro + 6 Thunderbolt to Ethernet Adapters = Fastest Way to Image?

[Mumbler]

The new MacPro has two ethernet ports. 6 more could be added with thunderbolt adapters, for a total of 8 gigabit ports.

For a while, OS X has provided a way to use several ethernet adapters with one IP address (http://support.apple.com/kb/PH14045).

With the high speed SSD in the MacPro, could I really get 800MB/sec output to the network with this setup, provided my switches supports it? I don't want to spent $3,000.00 USD on a MacPro and have this not work.

I would like to use this setup for both NetBoot and a Distribution Point. It would be used to image about 1,000 systems a year.

If this will not work for some reason, the 6 thunderbolt ports could still be useful for Target Mode imaging. I can already image MacBook Airs over thunderbolt in under 3 minutes, so doing 6 at a time this way would be pretty cool.

Thanks.

[alexjdale]

My two cents, you are adding a not insignificant amount of complexity with this, I wouldn't do it for anything more than a POC. I wouldn't rely on this in a production environment.

Also, I don't know how you can image more than one system at a time with target mode imaging? Aren't you limited by one instance of the Casper Imaging app on the Mac Pro?

[dwandro92]

I haven't tried imaging multiple computers at once, but I just ran a test and was able to open multiple instances of Casper Imaging using the following method:

1. Open Terminal and login is root via "su" or "sudo su"
2. Run the following command: ``` /Applications/Casper Suite/Casper Imaging.app/Contents/MacOS/Casper Imaging & ```
3. Enter your JSS URL if prompted and login
4. At the terminal, press Ctrl+C to clear the log messages that resulted from opening Casper Imaging and return to a shell prompt
5. Repeat steps 2-4 until the desired number of Casper Imaging instances are open

[monogrant]

If you create an aggregate with the two onboard NICs you'd be just fine. You will need a switch that support LACP link aggregation as well for this. Currently the largest LAG I'm using contains 4 members - 2 onboard and 2 from a PCI NIC in an Xserve for link redundancy more than speed.

In the end, your potential bottleneck on the Mac Pro won't be the network connection between the server and switch. You'll tap out your disk IO (even with the 800MB/sec flash) before you saturate the network connection.

Remember for the distribution point, the client will copy the file down and then run it's installer locally. Depending on the clients, you might have plenty of time spent with the client CPU/disk slowing the process while they install each policy/package.

You may have better luck running a similar setup to the setup Marco demonstrated from Oxford. Get a simple OS install and have it run a "first run" type of script at reboot to then have the client pull down and install the "SystemSetup" type of event policies on its own.

I've had far better luck with this setup than attempting to get all my packages installed at image time. It takes load off the net boot system quickly - the block copy OS install takes literally 3-5 minutes - and distributes the install processes onto the client systems. With this setup there is also almost no limit on the number of systems you can image simultaneously. As long a systems fit on your subnet/switch you're golden.

https://www.youtube.com/watch?v=Ka-OT17Mg-A
https://github.com/ox-it/mac-scripts

[activitymonitor]

This would be sweet if it worked.

@monogrant
If I recall correctly, Casper Imaging does not copy DMGs before installing them. They either get block copied with ASR or mounted and then the contents rsynced to the local drive directly from the server.

[mcrispin]

This is interesting theory crafting, but I don't think you've defined your functional requirements very well.

You said 1000 deployments per year. How many at once? What is the size of the image you are moving? And where is netboot in all this, or not? How fast is your turnaround for delivery? How many people do you have? Do you even have the space to image hundreds at a time, or are we talking about taking over a gym for a week here? Can the network support multicast? Or is this closed-loop? How fast do you need the turnaround? What is the post-delivery config?

It is not so much an image takes X time to complete, it also matter how many at the same time. Meaning - an overnight session that demands zero-touch but takes longer is more advantageous than a complicated but fast process that is bandwidth limited to just a few machines at a time.

What is your end goal? Then.. do the math.

Besides... Thunderbolt 2 won't just be limited to the Mac Pro soon enough. HINT.

[Mumbler]

I have 5 schools, each with about 200 student machines. I would like to spend a day at each building with this setup and reimage everything before the buildings get locked down for the summer. If this works as well as I hope, I would do the same over winter break.

Images are 10-15gb. Everything else is done with policies after imaging. Multicast is not an option on our network unfortunately. So this would require transferring about 3,000gb per building, plus the overhead for the netboot environment, which I have observed as 1-2gb per system, depending on the OS.

[bentoms]

Target Disk Mode imaging? As long as you can get all the macs in a room.

[John_Wetter]

I'd actually look at going the route of a Thunderbolt 2 to 10Gbps instead. Here's one example:http://www.attotech.com/products/family.php?id=15

[John_Wetter]

For those following via email, I found a much better Thunderbolt 2 to 10Gig ethernet solution. I edited my original post.
http://www.attotech.com/products/family.php?id=15

[Chris_Hafner]

If you're trying to save money you could buy a few external SSDs and use DeployStudio to deploy an image, which can then be recon'd. I've also had success in setting up external drives as fully bootable Distribution Points for mobile Pre-Staged imaging. Though, each needed a qualified DNS address, static IP and a copy of OS X Server running on it. Actually, at the time I used Casper Imaging to deploy a small fleet of them just to see if it was possible. Turns out that with the proper planning (and licensing) you could pretty easily deploy a small fleet of these and image vast quantities of machines all over the place with almost no bandwidth utilisation. Yet...

I'd be pushing to get some sort of simple NetBoot/DP(JDS) in each school. You probably wouldn't want to spend big money on a bunch of Mac pro's though. Some Mac Mini's could do the trick nicely. 200 units per school isn't that much in the grand scheme of things. With a MacMini NetBoot/JDS in each school you could probably avoid having to go to those schools in the first place. That's the beauty of Casper.

So, say < $5k and then no more driving/imaging outside of the repair process.

P.S> Yes, you can get close to the max speed of the Mac Pros drives as they are still far slower than either ethernet OR thunderbolt. Our fast imaging environment consists of two identical Mac Pros (SSD boot drives, Areca hardware RAID running 3 15k7 cheetah SAS drives, striped 0.) One as a primary Netboot and Secondary DP, the other the Primary DP and Secondary DP. Each set to fail over to the other. I've never seen the primary DPs network output as high as 450mb/s while imaging ~50 at a time.

Also, multicast is evil. Well, at least that's what I hear the network folks mumbling about ;-)

[mcrispin]

Can you say what your actual goal here is without using technical terms?

I WANT TO IMAGE ALL THE THINGS!!!

The whole concept of justifying a $4k Mac Pro with flipping adapters to solve whatever thing you have is just bad engineering.

I just bought a Mac Pro for my environment and it's going to become the most awesome-sauce VM shizznit evah, a much more appropriate task. The idea you are going to take that precious thing of engineering beauty and make it your traveling biotch makes me have sad face. I am going to kill every last crap windows box in my house because of my Mac Pro and you want to make yours the deployment whore. Jesus and Steve will be sad if you choose unwisely.

I came here for the experimental Mac Pro talk which got me all hot and bothered because Mac Pro has all the awesome things, and what I am getting is basic theory craft for a dude who pretty much wants to image about 200 machines per day over the span of a week with no help in 5 different buildings, which I can pretty much do out the back of my 1996 Toyota Camry with 5 mac minis and a nice Cisco 3000 series gig switch using pre-stage imaging and none of this crazy fancy stuff the rest of the crazy talk on this thread are suggesting. Multicasting? Really? Come. On. Thing is... you can repurpose all those Mac Minis... the Mac Pro... not so much... les we make Steve sad.

You don't have a technical issue dude, you have a workflow issue.

Stop talking to people about what you "think" the solution is and being all balls hard, and start being honest about your actual functional requirements to make your life easier and make you look like a god to your superiors.

If this somehow gets your people to buy you a Mac Pro - than awesome-sauce, but ain't no Mac Pro gonna solve this unless you can actually describe and implement.

Do better. Seriously. This ain't all that complicated. Segment for Christ's sake. Not everything has to do everything all the time.

"What is your end goal? Then.. do the math." ffs.

[Mumbler]

@mcrispin

Yes, 5 Mac Minis with SSDs would work almost as well. They would also cost MORE then a Mac Pro. I did do the math, and nobody has suggested a cheaper option to image 200 machines in 8 hours. I am open to suggestions.

[John_Wetter]

@mcrispin - As Nadeen at Burger Castle would say... "Simma Down Now!"

Yes, multicast is evil and people need to get over it. But, it's not a bad thing to want to image fast, and the way to do that is lots of flash drives and big pipes. That request doesn't seem unreasonable to me. Netboot is a drain, and moving files fast for imaging is nice. For new computers, your workflow will likely get hung up with unboxing, etc. so there are certainly workflow issues. But, for existing computers, you just want to crunch through them. If you've got the infrastructure and the people to do the imaging as fast as you can move them through, then go for it.

[Chris_Hafner]

This has certainly been an interesting post. I'm wondering if some form of adult beverage could be credited, at least partially, for some of the wittier comments here ;-)

In regards to cheap/fast imaging: Yep, external -bootable SSDs so say, 8 http://eshop.macsales.com/search/MBR2012A.480 @ $339 = 2,712 and are a heck of a lot easier to carry around instead of a Mac Pro. They would be about $80 Cheaper than the stock MacPro, far simpler and probably a bit faster as well.

As for the Mac Mini's I'm not trying to nit pick, but I probably would avoid running SSDs in the Mac mini's at least for the price at this moment in history. I would just grab 5 base mac mini's (which price out to ~$279 more than a single base MacPro, for example).

While they wouldn't be nearly as fast as the external SSDs for direct imaging, you could be continuously updating and/or imaging across all five schools at the same time. Overnight with any luck. After all, what's the difference between 5min and 60-120 min if no one is there to care.

So seeing as I'm not familiar with all the details of your specific situation, you may have some great reasons why everything I or we, mentioned would never work for you. I'd be happy to chat about your situation directly if you're interested. I am Instructional Technology Coordinator (Casper Admin amongst my varied responsibilities) for Brewster Academy. We've been 1:1 (Mac Laptops) for 21 years and BYOD for 4. We've hit about 90% of every technology issue that's ever existed in education and often consult with schools and school districts worldwide.

Regardless, there are many many ways to skin this particular cat.

P.S. Some numbers for some of the suggestions I've seen here.

• 8 OWC external SSD (480 Gig) = $2712 + Shipping (I picked 8 just because. You could easily do 4 for half the cost). These also come with great warranties unlike the MacPro or Mac Mini suggestions.
• 1 Mac Pro (base - 256 Gig SSD) = $2799 + Shipping + whatever adaptors you wish to use
• 5 Mac mini's = $2,895 + shipping

[Chris_Hafner]

@mcrispin if you want to geek out about MacPro imaging solutions I'm game. Perhaps in another thread!

[musat]

I just thought I'd post here that yesterday we imaged about 150 MBA in about 8 hours using just three Mac Mini servers, no SSD drives. Each Mac Mini was connected to the network using just the single gigabit jack. We are imaging 12 MBA at a time, with four going to each of the Mac Minis. We have netboot running on each Mac Mini, as well as having each being a Casper Distribution server. We have compiled our Casper image, and it is just under 15GB.

This process could probably be sped up even more by eliminating the netboot and booting via a USB 3.0 thumbdrive.

So very doable to get 200 done in a day without crazy configs.

[Chris_Hafner]

Nice... however, why so few on each MacMini? I know places booting 1-2 dozen netbooted from a single mac mini.

[musat]

Purely for netboot reasons. We found the netboot time increased dramatically as we added devices. Plus the shelving racks we have will hold four MBA. So, with one person in the room, we can pretty much keep a steady flow of MBAs going through the process. The imaging time increases linearly as we add devices. Meaning imaging 4 MBA takes four times as long as imaging one MBA. However, the netboot time increases more. So netbooting 8 MBA from one server may take 3 times as long as booting 4 MBA. That's where, if you could boot via some other method, you could do even more.

[Chris_Hafner]

@musat Fair enough. So long as it works it works. However, that is really poor netboot performance. You should see a slowdown but not that much unless you're running out of "space" for the user shadow files. If you ever want to take a look at that I know you can get far better performance. On far older NetBoot servers we wouldn't see any slowdowns until at least 10 netbooted computers at once. That was all the way back in the 10.4 / NetRestore days.

[mcrispin]

Hi all -- really solid responses all around, and lots of great ideas -- and yes, @Chris_Hafner I would *lurve* to geek out on MacPro imaging solutions, of which I am building one, but it is *very* site specific. I did get my vapors up because I really love the new MacPro because I think it is very new and different way of looking at desktops, I just hate to see that sexy beast shoe-horned into something that isn't appropriate for the MacPro's actual strength. I'm mean.. my god... all that Thunderbolt 2 + 1.4 HDMI... YIKES. How is this not the world's greatest information multi-display hardcore rad thing evah? I could run Chicago O'Hare with that thing, if only the cabling limitations weren't so.

Depending on your infrastructure and needs, something fun to try is dropping the base OS via netboot, and then using triggers post-restart for everything else for a thin imaging solution and cycling the machine through. It's true - the inventory, labeling and un-boxing stuff *always* takes more time than you'd expect. Nevertheless, if you run it like a factory chain, you can keep in constant motion and rip through hundreds of machines in very reasonable time and keep your feet moving and still have QoS. There is some awesome-sauce too to be leveraged using that Patchoo! thing so you don't get tripped up on patching.

I think if you are creative and have a grip on the tech, you can image far more than 200 a day with less than what I said, and remain mobile and cost-effective. All this is just *very* environmentally specific and you gotta deal with your local considerations. I think this is mostly why everyone always talks like "their" solution is the hot sauce, but doesn't translate well across different set-ups.

[tlarkin]

Hi Everyone,

Got a quick break here, and wanted to chime in. This is my opinion, from my experience, from supporting Macs for a long time now. Imaging can be a daunting task and it has many different methods and workflows. Some workflows are awesome, some are a waste of time. This highly depends on your situation and results vary depending on your end goal and environment. You have 4 options with Casper when it comes to imaging, which are:

1 - Netboot/NetInstall
2 - Target Disk Mode Imaging (thunderbolt/firewire)
3 - Local disk (external hard drive, restore partition, etc)
4 - Thin Imaging or provisioning

Each method has it's proper place. Best practices and workflows will vary from method to method. This is highly dependent on your environment too. Many things outside of Casper to consider, such as: Spanning Tree, storm control, IP helpers, and every other network infrastructure configuration that can affect Netboot.

As for Thunderbolt imaging. It is insanely fast. I have done POCs with it and customers over the past couple of years. I have deployed it many different ways too. Local SSD drives and Target Disk Mode. I have also done testing with many different image sizes, anywhere form 5 to 80 gigs, and I was always able to image a system over TB in under 5 minutes (including boot times) in that image size range. I have also used Casper Imaging and asr scripts, and they are both very fast. I have never used this in production though, only POC. Apple actually has a white paper on this:

https://www.apple.com/education/docs/Apple-ThunderboltWhitePaper.pdf

For what it is worth, I have mass imaged Mac laptops a lot, and myself and my former co-workers (at my previous job) imaged over 6,000 Macs in about a 2 week period. However, a lot of that time was logistics and not completely imaging. We had to unbox them, organize them, glue our asset tags on them, box them back up into groups so trucks could pick them up and drop them off at specific buildings, etc. We also could not use prestage imaging because of our naming convention we used. So, all Macs had to be manually named, but there are other ways to automate that and I just never had the time to build a solution for that. I used 3 Mac Minis, with 24 port gig switches at 3 separate imaging stations. We were a crew of about 5 people, who would sometimes get help from other employees at our org. So, at max, probably 10 people at a time working on our imaging projects. We did an assembly line style of grab boxes off pallet, unbox them, attach asset tag, netboot and image, box back up, and then organize them on pallets for pick up and delivery. I think our record was just under 600 Macs in one day. We also ran into network issues which slowed down or stopped Netboot at times.

As for multicast, it is something that can be done, but I honestly don't think it is worth the effort. As I understand it, and please correct me if I am wrong, is that multicast on other platforms is more intelligent and dynamic. Meaning, that it can dynamically change configurations to meet optimal settings. Where asr, when put into multicast mode is done so via a script (or how you start asr basically) and it is static. So, you have to figure out all the sweet spots for the configurations to make multicast to work well on the Apple platform. I never setup multicast ever in a production environment, not even back in the NetRestore (from Bombich) days. I did build other multicast solutions on other platforms. I POC'd a product called DRBL years ago (diskless remote boot Linux) and tested it out for massively imaging Windows PCs. I don't think multicast really ever made that much of a difference versus unicast. Maybe that is just my experience, and perhaps others have highly benefited from multicast, but I honestly don't think it really matters for most imaging workflows.

So, in conclusion, to my giant wall of text here is the TL;DR - Imaging has many different workflows and options. The best options will vary from situation and environment. $3000.00 for a POC, in my opinion, is most likely just a drop in the bucket in the grand scheme of things. I think Thunderbolt imaging is awesome, but I would be skeptical to say it scales properly in all directions just yet. However, if you follow that Apple whitepaper you may have awesome results.