a BIG INTRO And business Case for FreeNAS

[m1013828]

Hi everyone, I am working for a Laptop repair company in Shakeytown, New Zealand, (aka christchurch)
Iv'e been working there a year and have finally decided enough is enough and am trying to push through some much needed updates/upgrades
Currently we do a few brands of warranty repair, and when a hard drive fails, we burn off the 3-4 ISOs and do a manual install, which takes 4-5 hours per machine. This is becoming difficult as optical drives are being phased out, and the transfer speeds are horrible.
We have a few images for business clients using acronis that we have on a windows xp file server, and it also stores customer backups, there is no (gasp) raid or ECC ram etc on the file server.......
For corporate images, we remove the HDD from the laptop, and connect it via usb 2 to the file server and acronis the image accross, if the laptop is hard to access the HDD, we acronis to a spare drive, then connect the spare drive to the corporate machine, and boot with an acronis cd, and clone it AGAIN to the internal HDD.
The other staff seem to do things the way they've always been done, and it is very time inefficient................

My solution consists of 2 parts. Im deploying a PXE server that hosts ISOs only, (fog project) this allows us to load any ISO over the network, eliminating the need to burn discs, saving a little coin, and a bit of time either burning, or looking for discs (post quake we have been in a temporary location, so organisation isn't great.

The second part is a NAS, My vision is for laptops to boot into Acronis over the network, and backup or recover from images over the network, no removing hdds from machines, no usb2 junk, no bull****

for every SKU of laptop we can have several images, oldermachines have 32 bit and 64 bit Vista and maybe an XP downgrade image, windows 8 will be the same I suspect.

we will literally have hundreds of acronis tib files on the NAS, currently there is no protection, cost is a factor, so we wont be shelling out for a windows server license.. Plus I've wanted to deploy ZFS for my own professional developmentfor some time.

So here are my questions.
Data deduplication We will have hundreds of very similar, large files. The documentation is inconsistent about how much ram is needed for data deduplication. Ive read 1, 5 and 8gb per TB of storage. What is the communities' consensus on ram/storage ration? I have to take cost into consideration, at some point Ram+ dedup will become cheaper than throwing HDDs at the problem, so I want to consider this at the outset.

SSD ZIL. we currently have 4 technicians, down from 8 before the quake, but we will be building up our numbers again. As a worst case scenario, we could have 3 technicians writing 30gb Images to the NAS at the same time, provided the NAS has enough bandwidth to service 3 clients fully, will an SSD ZIL help performance?? I see lots of references to database apps, but we have huge files, so I'm assuming huge sequential writes, so I'm hoping I can make a business case for SSD ZIL.

SSD L2ARC, considering we have massive images, would L2ARC be worth while, again, worst case scenario is 3 sets of 30gb being pulled off the NAS at the same time, I want this system to last for years so I'm keen to reduce disk thrashing. I was thinking maybe some of that data that would normally be earmarked for deduplication could end up on the L2ARC??

Proposed setup,
I have considered my options, and think that a Consumer mobo and CPU should suffice, so long as the mobo has 8 SATA ports, I would employ 6 Western Digital 2TB Reds in RaidZ2 giving us 8tb of storage. Also would be included is 2 SSDs, one for L2ARC, and one for ZIL. Ram would be at least 16gb, and as much as I can fit if Data deduplication is viable. As we would be using CIFS/SAMBA shares I wouldn't skimp on a low end CPU either.
Network would consist of a Intel Quad gigabit NIC bonded, allowing the effective servicing of 4 clients at once.

Are there any pointers on the proposed hardware??

Scrubbing, despite being prosumer level drives, I would probably setup up saturday night scrubs weekly. How long can this take with say 8Tb of data??

Anyother pointers??

Whew!!! what a big intro:)

 

[cyberjock]

Deduplication = BAD IDEA. Especially since cost is a factor. Just stick with compression(which I think is pointless for already compressed images). Consider this. If an image is 20GB, you'd need 100 of those to equal 2TB. 20GB is a bit big from personal experience, and a 2TB drive can be purchased for less than $100 on sale. Trying to save money by using deduplication is just a waste of your time. If your system has an unclean shutdown you may not be able to get access to the zpool until you get enough RAM, and since I'm sure you don't want to run out and buy 32-64GB of RAM and find out how much you'd need, just stick with compression. Deduplication also works at the block level. While your images will have a large number of files shared it's likely that the compression will screw you over and yield minimal savings if you use deduplication. That's why the release notes for 8.3 say:

***** - ZFS v28 includes deduplication, which can be enabled at the dataset level. The more data you write to a deduplicated volume the more memory it requires, and there is no upper bound on this. When the system starts storing the dedup tables on disk because they no longer fit in RAM, performance craters. There is no way to undedup data once it is deduplicated, simply switching dedup off has NO AFFECT on the existing data. Furthermore, importing an unclean pool can require between 3-5GB of RAM per TB of deduped data, and if the system doesn't have the needed RAM it will panic, with the only solution being adding more RAM or recreating the pool. Think carefully before enabling dedup! Then after thinking about it use compression instead. *****

ZIL and L2ARC are probably pointless for your function. Read up on what a "sync write" is and you'll understand why ZIL is pointless. You're going to write files once in a while, but do far more reading. Why add a ZIL to a system that will have alot of writing intially during your fill-out stage, then almost no writing after that? For L2ARC you aren't looking to saturate Gb LAN so I see no value added(see below for speed estimations). Save your money and buy a spare 2TB drive. Then one one fails you have a spare ready to go. The scrub duration is dependent on the speed of your hard drives and CPU. Buy a Celeron and expect a filled 8TB zpool to take 12-24 hours. My low end first gen i3 can smoke through 15TB of data in about 9 hours. I always recommend a high frequency i3 if you use CIFS since it's single threaded. They're also fairly cheap so you'll save a little money there.

A 4 port gigabit NIC is total overkill. Even with Gb, you'd be lucky if 4 machines used 100MB/sec running simultaneously and combined. Because of the decompression and writing speeds of the data in Acronis you're not exactly going to have a need for more than 1Gb LAN port unless you plan to do more than 4 machines simultaneously ALL of the time. If you have all SSDs in your machines then you MIGHT be able to max out a 1Gb NIC with 3-4 machines simultaneously. Acronis is just too slow when it comes to decompressing and writing data to the drives. If you want to go "crazy overboard" get a dual Gb NIC. They are pretty cheap if you feel compelled to spend the money. It's not like you can't upgrade to a dual port(or quad) later if you find yourself using 90% of the bandwidth of a single Gb LAN port.

Think a little smaller.. you aren't running a Dell factory where you are imaging 100s of drives a day. The bigger you build it the more stuff that can go wrong. It would totally suck to overdesign your server and you end up spending more time than you save just keeping the server up. Remember KISS(keep it simple stupid). You only have 4 techs, but even with 8(you said you used to have 8) I can't imagine all 8 of you sitting around trying to image 8 workstations all simultaneously all of the time.

 

[m1013828]

Thankyou for the awesome reply.. you shattered my dreams!!!!!!! lol. I can take the crticism.....
Okay, no dedup, not worth the hassle if acronis compression screws with the potential savings, costs and risks > gains.

no ZIL, would only be usefull during the initial construction of our image library. After the library is up to date, we will only have a few manufacturers images added per month. Though maybe a few customer data backups per week, which will only be stored for a month or so.

SSD L2ARC, may do anyway, I want this server to last many years, and the client machines will only become faster and faster. the business cycle is 3 years for HP stuff, so i'd hope for it to last 6 years before needing to be replaced.

Network speed, I obviously have to test the Reimage speed of Acronis over a gigabit network to test its efficiency, it is contrained by a 100mb switch at the moment, so I will have to look into that, I was under the assumption I could essentially stream the images as fast as the NIC will allow, (60-75meg a second for realtek etc, and 90 for intel) as the HDDs are faster than the NICs in most new machines. If Acronis does have a huge overhead I will try and find another imaging solution that is faster over a network. We are essentially starting from scratch, so Im not particularly wedded to Acronis, it's just what's been used previously. Time taken may seem trivial, but it reduces our workload of concurrent activities, and improves customer satisfaction. I was working on 2 gigabit ports would service 3 clients at max speed. Bearing in mind that we may have several imaging jobs per technician concurrently I often have have run out service desktops to run backups, scans and restores on..

I want this setup to be a over engineered, but not overkill, It needs to just work, for years on end, repairing corruptions, unlike the old win XP server. ( XP and server in the same sentence yuck!)

 

[cyberjock]

Network speed, I obviously have to test the Reimage speed of Acronis over a gigabit network to test its efficiency, it is contrained by a 100mb switch at the moment, so I will have to look into that, I was under the assumption I could essentially stream the images as fast as the NIC will allow, (60-75meg a second for realtek etc, and 90 for intel) as the HDDs are faster than the NICs in most new machines. If Acronis does have a huge overhead I will try and find another imaging solution that is faster over a network. We are essentially starting from scratch, so Im not particularly wedded to Acronis, it's just what's been used previously. Time taken may seem trivial, but it reduces our workload of concurrent activities, and improves customer satisfaction. I was working on 2 gigabit ports would service 3 clients at max speed. Bearing in mind that we may have several imaging jobs per technician concurrently I often have have run out service desktops to run backups, scans and restores on..

I want this setup to be a over engineered, but not overkill, It needs to just work, for years on end, repairing corruptions, unlike the old win XP server. ( XP and server in the same sentence yuck!)

Honestly, I've played around with a few other programs, and they all have similar timetables for recovering data. The issue is that you are trying to recover tons of small files. That means lots of writes all over the place. I can image slow laptops with SSDs in record times compared to high powered workstations with platter drives. The limitations are almost certainly the seek times for platter drives(thats what makes SSDs so awesome?!) so regardless of what program you use you'll have the same limitation.

When it comes to ZIL and L2ARC, it's not like you can't add it later, right? I'd definitely try using the zpool without anything added just to see what the performance is. If you really think that the server is being held back then try an L2ARC. L2ARCs are great for situations where you re-read the same data over and over. But you won't be doing that, so the performance benefit for L2ARC is just not there. You'll use 1 image today, and more than likely a totally different image tomorrow. Even if you use that same image every day for a week, you're still talking about not saturating Gb speeds. If you think that your zpool won't be able to send data at 100MB/sec then you have bigger problems :P

 

[m1013828]

So now that you've cut my ambition down to size, Im going to benchmark network utilization and time a typical image transfer on a gigabit network segment, to see what network connectivity I will need. I may just find a mini ITX system with Dual NIC and 6 Sata that sounds like a good mid term solution, and wait and see if the Network gets saturated.

 

[m1013828]

We have an old "big Iron" server from before the quake with 4 scsi disks, dual socket etc, a really old heavy dunger. I will set this up with freenas to benchmark, our alternatives are to get Linux Acronis boot disks, and backup/restore over nfs, Acronis boot over cifs, and clonezilla over nfs, I will benchmark these with a large image to clone, and monitor throughput etc to plan for the real deal.

Recommended CPU, you suggest just a fast dual core? as CIFS isnt multi threaded? or would a quad core help for multiple concurrent NFS transfers? I was thinking of slightly undervolting an i3 3240, and try and fit it all in the Fractal array R2 case, (with a faster case fan than the stock supplied one, I hear the 600rpm fan is a let down on that model)

 

[cyberjock]

For CPU you want at least 2 cores, but you want more power per core versus more cores.