Guide To Building A Media Storage Server - Page 152

If you are dropping UnRAID as your OS you don't need a board with PCI-X slots to use the older SATA cards. The board you have will work, but I would go with something with a little more PCIe expandability if I was you.

What OS do you intend to use? And what RAID system/type? Really you need to nail down exactly what it is you want to do with your hardware before you pick what that hardware is going to be.

I thought I will get a board that may go PCI-X *and* PCI-E like the one above.

The Supermicro X7SBE would allow me to use two PCI-E (4x) based AOC-SASLP-MV8 and also reuse some of my older PCI TX4's but also use the PCI-X based AOC-SAT2-MV8. It would be unRAID and FlexRAID/WHS compatible, even Linux.

However, the more I read about FlexRAID/WHS the more conviced I am.

I absolutely don't like the performance limitation of unRAID!!

Fair enough. As that board has dual, independant PCI buses you could even try your TX4's in it and see if you like the performance. By having 2 buses you instantly double the available bandwidth for each of the drives attached to your PCI cards. The 6 onboard ports, plus 2 from each of 2 TX4's (one on each bus), would give you your 10 starting ports, and you save the expense of buying new controllers that you don't (yet) need.

Sounds absolutely reasonable! Thanks much!

Anyone using remote management with Supermicro boards? The AOC-SIMLC is available for the X7SBE. Sounds good to be able to setup BIOS or having the serial console or KVM over the network, be able to reboot, etc. Any thoughts?

Thanks guys for all the unRAID/hardward/speed responses. It was very helpful. Regarding performance, I don't want to build something and be too limited. IOW, I want it to perform. But as you guys all know, there are trade-offs for performance, cost (which has many categories), ease to maintain, etc.

As I've said, I don't know a thing about FlexRAID. Sounds like it works with WHS, I'm definitely going to have to read up on it.

The one thing I liked about unRAID was no hard drive was needed for the OS. I'm also a little leery of using a MS OS. But it seems like WHS has lots of followers. Hmm, more to think about it.

Starcat, any chance of you doing a thread to document your build? I'd love to steal your work!

If you are thinking about FlexRAID then take a look at disParity. It's a simple line command and no bloated installation. Does exactly what FlexRAID does but a whole lot easier to use since there is only like 4 or 5 commands. I've tested it with a full 1.5TB drive recovery and it works.

ejhuzy, thought about this, but it was too late to start a new thread as I already have posted here in this thread. Any chance the admins moving this posts to a separate thread? I will of course continue later on and document with pics, etc.

ejhuzy, you can use FlexRAID with Windows and Linux, it is not limited to WHS only. Take a look here. I like being able to insert drives already containing data without the need to reformat/init them (filesystem must be native to the OS however) and only calculating parity info.

It might not at the beginning when the file system is empty and block allocation happens sequentially but if you start deleting things (e.g. with PVR recording), you will create fragmentation and some performance loss over time.

I think the real myth may be this . Let's assume you have 100 files per BD/DVD title. If you have 1000 movies stored on your array, that is 100,000 files. Assuming we wasted half a cluster on the average, that is 32,000*100,000 = 3.2 Gigabytes. For BD, storing 1000 titles will take an array that is probably 35 TBytes or 35,000 gigabytes. 3.2 Gigabytes would be round off error in that. Double or triple that and it still means very little. Considering that you always have some loss even with smaller cluster sizes, the penalty is actually less than it seems.

Also, keep in mind that small files (less than 1K) get stored in the MFT directly and don't use up a cluster at all. So the actual penalty will likely be even less.

Since there is nothing but upside for larger clusters, I think in this application it makes a lot of sense to use larger sizes. With the larger clusters, you are assured that there is a nice floor to fragmentation of the drive.

If you aren't going to use unRAID and be shackled to a PCI-X card, it is pointless to waste valueable slots on obsolete technology. I would suggest you consider a board with PCIe slots.

They provide great features, and are invaluable in an enterprise environment.

Whether they are useful in the home or not depends on your own situation though - the cards are expensive, so unless it is going to be a real pain it might be worth putting up with just using RDP/VNC to manage the server while in an OS, and setting up a local terminal if you need to get into the bios, reinstall the OS, etc.

For most users, once you have set up the server you don't need to touch it again, so the cost of the management cards is not really justifiable. If the server is in up in the roof or something they may be advantageous, but I would imagine that is more likely to be the exception...

So this it the list of parts that I have put together for my 10tb mid-tower server project. I won't start out with all the hard drives of course as I'm waiting for 2tb hdd prices to drop.

Already Have:
WD 120gb 2.5 Scorpio HDD
LIAN LI Lancool PC-K7B
PC Power and Cooling Silencer PPCS370X 370W

Planned:
ASUS M3A76-CM AM2+/AM2 AMD 760G Micro ATX
AMD Athlon 64 LE-1640 Lima 2.7GHz Socket AM2 45W Single-Core Processor
Western Digital Caviar Green WD20EADS 2TB SATA 3.0Gb
Windows Home Server

Possible CPU Coolers (will be passive)
ZEROtherm BTF95 CPU Cooler
Scythe SCMNJ-1100

I also have 3 quiet 120mm fans for the case for airflow. I have used this case, PSU, the Minja and fans in another build for my brother and liked the look and results. I plan to add more hard drives later of course.

This will be connected to my HTPC via that Cat5 cable, I guess. I no nothing of RAIDs and I plan to rip media on the HTPC and transfer to the server. I use SageTV, PDVD, My Movies, Media Browser. I will record the OTA HDTV shows via 2 Avermedia Duets. The cards will be in the HTPC and I guess I'll save the files to the server. Or I may keep the 1tb drive in the HTPC and record the current season there.

What to do guys think of the shopping list for a first time server build?

I was looking again on the Supermicro site and just can't find a "better" mainboard than the X7SBE. Others don't really have much more PCI-E ports and always have a couple of shared PCI slots. The X7SBE has the same PCI-E but adds PCI-X 133Mhz/64bit to plain PCI. PCI-X might be "obsolete" but is simply fast and is backward compatible to PCI (two separate busses). Or do you have any specific example for a mainboard I might consider for a 4220 based media server?

I might go cheaper, well, but I very much like the IPMI service processor which I will be adding just to not have to bother with keyboard/mouse/monitor when installing or configuring the BIOS. It will also get me the serial OS console over ehternet and just make life easier.

If you are building based on parts in stock, your arguments are sound. Or perhaps even if you are building a 2nd/3rd/etc system and needed to match parts from your prior builds in order to manage spares/repairs.

But if you are building from scratch and buying new components for your build your arguments make no sense at all. You simply cannot match PCIe performance using PCI-X parts and the total cost of building using PCI-X will be higher.

Basically, it comes down to a price/performance ratio issue sweetened with a touch of limiting future mod options.

Buying used / eBay puts you somewhere in between these extremes, but most likely price/performace would still tip towards the PCIe based build, ultimately depending upon your success "shopping victoriously".

Piglover, I am building from scratch. I was trying to make a system that will do it all, being able to run unRAID, WHS/FlexRAID, Linux, etc. However, as I am leaning more and more towards WHS/FlexRAID and two of the PCI-E (4x) based Supermicro AOC-SASLP-MV8 in addition to the mobo SATA ports, please feel free to sugest a PCI-E based/more modern board, preferably Intel based Supermicro. I would of course love to save a few bucks. Thanks again!

Tom has added support for this card in unRAID release 4.5-beta4. See this thread... however, see this thread too, looks quite problematic.

Could someone try to explain to me the difference between regular green power drives and their raid edition counterparts? I know the former lacks some features like TLER, but how critical are these features? I'm planning to use 20 of them in a raid6 array.

Anyone using the ARC-1680IX-24 24 Port Serial ATA/SAS RAID Controller? 5x of its port would make a great match for the 4220, the sixth port (and possible all from a second 1680) running to another 4220 holding disks only as a future expansion. Not cheap though and 10 GigE to the switch may become necessary but highest performance... A mainboard with 2-3 PCIe 8x slots to accomodate two Areca and 10GigE is quite easily available... Any thoughts?

On the downside *if* something happens to the controller/array, everything is gone at once!

From my research it appears that Linux doesn't support the Supermicro PCIe controller and that's a huge problem for unRAID. I'm sure after reading the unRAID forum exhaustively that Limeware is working hard to get it working (since users are clamoring for it) but it's not ready for prime time. I really hope they succeed but I wouldn't even try it without complete testing and support. The controller is not something you want to take chances with...

Do these (Discrete SATA to SFF-8087 Mini SAS Reverse breakout cable) work with the Supermicro AOC-SASLP-MV8? They are pretty cheap compared to the ones I have seen elsewhere.

Reverse breakout is for connecting discrete mobo or controller ports to a SFF-8087 backplane. So, no it is not appropriate for a AOC-SASLP-MV8. It would however be useful for connecting a AOC-SAT2-MV8 to a Norco 4220 backplane.

I'm kinda surprised you're asking because that web site you linked is pretty clear about what it's for.

- Mike

Wow. That seems like a bit of a turnaround. To start off looking for a cheap stacker case with 5-in-3s, then end up looking to buy 2 $1200 hardware RAID cards & chassis to match, connected up with 10gig pipes...

Honestly, all that is way overkill. There is absolutely no need for 10gig anything in the home - even agregated gigabit is way over the top 99% of the time. Even the Areca card you list is a bit over the top - a better and cheaper option would be a Highpoint RocketRAID 4350 and a Chenbro or HP expander to give you the extra ports - this is almost half the price, but gives you the same RAID processor, and IIRC Alamone says the combination works well. It would also be a much more elegant way of linking chassis together - a single multilane cable rather than running 5 cables.

For a user looking for a set-and-forget media server, I would be hesitant to recomend a controller until it is supported by the kernel, let alone one that is not even available in a production release of the software.

anyone..?

The RAID drives are apparently built to higher specifications, and as a result have longer MTBF ratings etc, and a longer warranty. They also often have more advanced features (vibration sensing etc). They are not really so useful in the home environment though, where the odd re-read will not cause any problems. The difference between a 3 year and a 5 year warranty is also not all that much - you will find it rare that a 4 year old drive dies. They tend to die young, or just keep going.

Overall the big difference is TLER as you mention. This is essential for running hardware RAID, as otherwise the drives can drop out randomly for no real reason. For WD drives it used to be a simple case of using the WD utility, but according to recent reports above that may no longer be possible.

I am not really sure what the best course of action is currently regarding this issue. I have not yet found any documentation regarding TLER-equivalent settings on Samsung or Seagate drives (although to be honest I am not in the market for drives at the moment, so haven't really looked into it). It may be that, at least for WD, it has to be RE drives for RAID arrays from now on, unless you take a chance on getting older drive models that can still be updated with the TLER utility.

Thanks, beast! Yeah, the stacker idea was a quick and dirty solution for the moment. I never really liked the idea, also the stackers are so expensive for what they are actually offering. Ventilation is a nightmare for the drives too. I couldn't find the Lian Li case (Lime Tech is using) anywhere, so I then found the cheap and good 4020/4220 cases and found out that unRAID don't offer the performance I need (at least sustaining 95MB/s over GigE which I am using). Money is usualy a concern but if I get an expandable and future-proof solution, I might spend well lot more, no problem here. So, being a Solaris guy I was always hoping to use ZFS, but due to poor HW support I kinda turned away from this idea, so was really surprised to see Areca supports Solaris. I wont run any applications on that file server. I would need NFS (AFP and SMB are not a must). Areca offer full Solaris support. However this is not a must, I may end up using WHS/FlexRAID, eventually Linux, but I would like easy admin and a HW RAID card is esentially on the easy side. A robust file system is important, the ability to spin down individual drives (and not the whole RAID volume as whole) is a huge plus. Read that Areca offers that functionality.

Comments above have turned me away from PCI-X which was mainly for unRAID compatibility which I am dropping as a requirement mainly because of poor performance!

I currently don't have a 10GigE switch and would not buy the 10G NIC immediately. It will come however as soon as I add more drives, expand to 2nd/3rd chassis and the need arise. I am currently running GigE ProCurve 1800 switches.

Will take a look at the Highpoint and Chenbro rigs. Please feel free to recommend any decent mainboard with 3x PCIe 8-lane slots, including IPMI service processor and decent Intel NIC/NICs. I think this leaves only Supermicro? Probably Intel, Chenbro too. Any HP part numbers for the expanders you mention?

If you like/want/feel comfortable with Solaris then stick with it. And if Areca support it all the better. But don't get a hardware RAID card - get the ARC-1300ix-16 non-RAID HBA instead. It will take 16 internal drives, with another 2 4-drive ports externally to loop through to another chassis if required later. Best of all it only needs 1 PCIe x4 slot, and costs about $360. I would assume Areca will have Solaris support for this range if they do so for their RAID HBAs. Then set up RAIDZ2 and you're away.

What kind of network infrastructure do you have? Why switches? Sounds impressive for the home. But still, you will never see any advantage from spending the cash on 10gig. If you want higher speeds to your workstation then maybe look at a dedicated gigabit subnet between it and your server, or even agregate 2 or more ports if your switch will support it. You will never see the benefit with media streaming alone. If you go 10gig you are just setting yourself up to spend a LOT of money just to get a headache and the same real-world performance.

If you are comfortable with 1 system, it can do everything you need and is cheaper, why learn something new? (at least for a production machine - feel free to play with the other stuff as you please... ).

You are absolutely right that ZFS do not need a RAID card. Will check out the ARC-1300ix-16 card and look for a mainboard with Solaris support. The Highpoint HBAs don't support Solaris.

I am running two HP ProCurve 1800-24G switches connected to patchpanels connecting all the wallsockets out there. No need for a dedicated link to my workstation as the switches can accomodate the bandwidth from all ports. I am happy with the performance. I currently only have the Qnap 509 which maxes out at about 80..85MB/s.

Ok on 10G. The idea come to mind for a single huge server doing more than 95MB/s which will saturate a single 1GigE link, however port aggregation may be used here to the switches.