Recommended NAS?

[mikesm]

Originally Posted by KJake

The cables the drives are using is fine, as they're all using the same cables and work fine for over a week not connected to the PMP. Connected them to the PMP and boom.

I should check the cable from the PMP to the controller, I don't think that I did that. I think that's about the only possibility left here.

The drives (500GB) that are connected to the PMP aren't in the 5-in-3, but I used 2 SATA power plugs to power it, direct from the PSU, not split or anything (since the 5-in-3 is going to "split" the power anyways).

Did you use all the plugs in the 5in3? Try using the 3 4 pin molex ones...

[KJake]

Originally Posted by mikesm

Did you use all the plugs in the 5in3? Try using the 3 4 pin molex ones...

I'm not using the 5-in-3 right now...these drives are connected directly to the PMP.

When I _did_ hookup the 5-in-3, everything was fine with it.

[unkyjoe]

Mikesm,

I am currently in the process of building my NAS as well.

I prefer Fedora Linux as this is what I have used in the past "a little linux experience" I am using version 8 but it does not have the EVMS software in it.

I am wondering if I would have to recompile me kernel to use this software to manage my drives, or does the Fedora kernel support it?

Or would you recommend I use Ubuntu or Suse?

I have not messed with debian before and I hate the cli so I want a gui on the server and the ability to connect from my laptop in the LR using NX client.

I am open to suggestions, "and a little hand holding"

Thanks in advance

[mikesm]

Originally Posted by unkyjoe

Mikesm,

I am currently in the process of building my NAS as well.

I prefer Fedora Linux as this is what I have used in the past "a little linux experience" I am using version 8 but it does not have the EVMS software in it.

I am wondering if I would have to recompile me kernel to use this software to manage my drives, or does the Fedora kernel support it?

Or would you recommend I use Ubuntu or Suse?

I have not messed with debian before and I hate the cli so I want a gui on the server and the ability to connect from my laptop in the LR using NX client.

I am open to suggestions, "and a little hand holding"

Thanks in advance

What kernel is the latest fedora shipping with? It may not have PMP support in it (I specicfically opened a request with the suse folks to make sure the PMP patch made it into 10.3, which wasn't hard since the author of the patch is a novell employee). If you are comfortable with patching kernels, then go ahead and use fedora if that's what you know.

You could install the evms rpm on fedora, as it's known to work there, but I believe fedora has a gui interface to lvm and raid, which is really all you need. I preferred the command line for setting things up, but to tell you the truth, I use the GUI utilities in Suse now anyway.

I'm always glad to help people join the "superfast N terabyte" club.

Thanks,
Mike

[unkyjoe]

Mikem,

Thanks for the help but after futzing around with Linux for 2 days I realized why I havent used it before.

I am a point and click guy, for one thing Samba never achieved the transfer rate my W2K3 server does, and yes I did all the tweaks.

Second thing I forgot since I am a re-seller I have a NFR copy of W2K3 I can use for myself. Had it setup in 2 hrs last night using Raid 1 to mirror my drives.

The thing screams! So I will stick with Windows for now, and revisit Linux when I have some more free time in my life.

Dont get me wrong I love Linux but out of the box I needed to get this up and running quickly without a bunch of tweaking.

I left the command line back with dos 6.22, except when I need to use it under windows to tweak some stuff.

Thanks for the help though.

[mikesm]

Originally Posted by unkyjoe

Mikem,

Thanks for the help but after futzing around with Linux for 2 days I realized why I havent used it before.

I am a point and click guy, for one thing Samba never achieved the transfer rate my W2K3 server does, and yes I did all the tweaks.

Second thing I forgot since I am a re-seller I have a NFR copy of W2K3 I can use for myself. Had it setup in 2 hrs last night using Raid 1 to mirror my drives.

The thing screams! So I will stick with Windows for now, and revisit Linux when I have some more free time in my life.

Dont get me wrong I love Linux but out of the box I needed to get this up and running quickly without a bunch of tweaking.

I left the command line back with dos 6.22, except when I need to use it under windows to tweak some stuff.

Thanks for the help though.

No problem. I always tell people to avoid Linux unless they are comfortable with unix OS's, otherwise it's better to go the windows route with a hardware controller. There some reports that windows server 2008 software RAID is MUCH MUCH better than 2003 software raid, which is good news. If you are going to a windows build, try and get a copy of that.

[davey_fl]

Admittedly I have not read the whole thread yet, but I'm wondering if anyone has tried the DNS-323 NAS box from D-Link? Currently it seems to be limited to either JBOD, RAID0 or RAID1. I'm concerned about the filesize limitations and the lack of NTFS support. Does anyone have experience with this box?

Thanks

[mikesm]

BTW, for those of you looking to build a really big NAS, an alternative to the classic stacker case has shown up - the Antec 1200. See some pics here: http://www.xtremesystems.org/forums/...d.php?t=146968

This case has 12 5.25" bays, capable of holding 4 5in3 SATA mobile racks, for 20 disks. Good ventilation, and two rear fans. Only one PSU space, but that should be fine if you are using staggered disk startup.

Nice to see more cases being aimed at the NAS sweet spot.

[BobPhoenix]

It's sold out but here is one that would hold 30 hdd's.

[briands]

Originally Posted by mikesm

BTW, for those of you looking to build a really big NAS, an alternative to the classic stacker case has shown up - the Antec 1200. See some pics here: http://www.xtremesystems.org/forums/...d.php?t=146968

This case has 12 5.25" bays, capable of holding 4 5in3 SATA mobile racks, for 20 disks. Good ventilation, and two rear fans. Only one PSU space, but that should be fine if you are using staggered disk startup.

Nice to see more cases being aimed at the NAS sweet spot.

Sorry if it's a dumb question... but what controls the staggered disk startup?

[mikesm]

Originally Posted by briands

Sorry if it's a dumb question... but what controls the staggered disk startup?

The disk controller. In the case of hardware raid, it's the driver for the raid controller. In the case of software raid, it's the SATA/PMP driver. FOr PMP connected ports under linux, the PMP driver expressly is coded to do staggered spinup. With windows drivers, who knows. I am pretty sure the Intel AHCI drivers don't do that nor do they expose that as an option to the user, but there could be some obscure registry hack that could enable it.

[Article22]

Wish I had found this post before I started building my own Linux NAS, 6 months ago...

I haven't played with LVM as I decided to stick with just a fixed raid 5 setup using mdadm but by the sounds of it - it is better to sort it out before you build your array

My problem currently is that while I used a similar 4 in 3 enclosure to allow for hotswapping sata drives (an IcyDock MB454SPF) I foolishly decided that I wanted a quieter system and removed the fan attached to the enclosure.

Needless to say eventually the drives all came up red and failed, since then while mdadm seems to be happy, SMART is emailing me after every reboot saying that the drives are unhappy (sector errors) though nothing has failed *yet*

Now any suggests on what I can do? Has anyone experienced something similar?

My thoughts are to remove 1 drive at a time using /hotremove, then install it, re-format it, check for errors, then rebuild it into the array - repeat for all 4 drives.

Would this help or are the drives actually about to die?

Only point I would make about doing your own NAS, is to use a disk array for your OS now if you can as it seems pretty pointless to have 2 drives hanging off connectors just to Raid 1 your OS...I have 4 400 Gb Samsung drives, and partitioned like thus :

md0 5 gb -> raid 1 using 2 drives as /
md1 5 gb -> raid 1 using 2 drives as swap
md2 1.2 tb -> raid 5 using 4 drives as /storage

It used to be very hard to set this up but now with the modern kernels and mdadm it is pretty trivial. I used a swap partition because you have to use the 5 gb off all the drives and you can only raid 1 (maybe 0) your OS still.

Thanks for your help,

Nigel.

[mikesm]

Nigel, if the drives have been subjected to continous overheating for a prolonged period of time, I would not use them in a system, particularly if they are showing lots and lots of smart errors. Heat is death to disk drives, and while those mobile racks are great in terms of density, they depend on forced air to keep things cool. Personally, I don't see how you could have removed the fan, as on all my docks, if the fan stops I get this very alarm that is hard to to miss.

Maybe if they are still under warranty you could get them RMA's by the vendor, but it sounds like it's only a matter of time before they completely fail.

Yes, with modern kernels and distros, you can certain run the system disk on the array as well, but I have to tell you, I don't think this is a good idea. Keeping the system disk separate simplifies a bunch of things, and if you have to move the array to another system, it's a lot easier to install a system on dedicated system disk first. Also, if the system disk blows up, it's still pretty easy to recover it from backup and keep going.

Also, from a performance perspective, in a RAID0 or RAID5 configuration, you want all the drives seeking to more or less the same place at the same time for maximum performance. If some disks are also hosting system partitions, and they are doing seeks and reads for that, but others are not, then the other disks are going to end up waiting when they could otherwise have fulfilled their requests and moved on to the next request. I don't think this is good from an IO perspective.

[mikesm]

BTW, frys.com has the WD 500GB SATA disk on sale now for $89.99 ( a new low in terms of $/GB) plus free shipping here: http://shop1.outpost.com/product/536...Homepage:Spot1

Of course, if you are building a BIG array, you should use denser drives, but I thought folks would like to know about this.

[Article22]

Thanks for that info Mike

Agree, it was very daft (you live and learn) to remove the fan but there wasn't any alarms when I did and it was only when it overheated did it start to squeal at me.

I think the array was only left on without the fan (there are two others in the case anyway) for a matter of hours (without much use) and when the alarm sounded I shut everything down and removed them immediately.

I was hoping that the errors were only caused by the overheating and a full re-format would stop the errors?

Thats very cheap for a HD! That would work out at ~£40!
The cheapest I have found over the pond is this :

500 GB Samsung HD501LJ Spinpoint -> £63.32

I am using the 400 GB equivalent and really rate the Spinpoints - very quiet and generally reliable...just don't fry them

Interested in your points on a disk array - while I haven't done much testing on the drive array itself, I have done a fair bit of tuning on the Samba side as this is all it is used for, and was quite happy with the 40-50 MB/s I am getting regularly over 2 trunked Gb ethernet ports, using a PCI SATA disk array (so the PCI bus is being hammered)

My method just means you can use a small stripe along the disk array for your OS and swap and still gain full 1 disk fail redundancy without putting in other specialised HDs (as I had no more ports!)

I agree that in a large array using PMPs it is far better to have them separate as the main cost in the system is the main drive array but in my case I wanted something reasonably cheap so I decided to use the main drive array for everything...

Cheers for your comments,

Nigel.

[mikesm]

Originally Posted by Article22

Thanks for that info Mike

Agree, it was very daft (you live and learn) to remove the fan but there wasn't any alarms when I did and it was only when it overheated did it start to squeal at me.

I think the array was only left on without the fan (there are two others in the case anyway) for a matter of hours (without much use) and when the alarm sounded I shut everything down and removed them immediately.

I was hoping that the errors were only caused by the overheating and a full re-format would stop the errors?

Thats very cheap for a HD! That would work out at ~£40!
The cheapest I have found over the pond is this :

500 GB Samsung HD501LJ Spinpoint -> £63.32

I am using the 400 GB equivalent and really rate the Spinpoints - very quiet and generally reliable...just don't fry them

Interested in your points on a disk array - while I haven't done much testing on the drive array itself, I have done a fair bit of tuning on the Samba side as this is all it is used for, and was quite happy with the 40-50 MB/s I am getting regularly over 2 trunked Gb ethernet ports, using a PCI SATA disk array (so the PCI bus is being hammered)

My method just means you can use a small stripe along the disk array for your OS and swap and still gain full 1 disk fail redundancy without putting in other specialised HDs (as I had no more ports!)

I agree that in a large array using PMPs it is far better to have them separate as the main cost in the system is the main drive array but in my case I wanted something reasonably cheap so I decided to use the main drive array for everything...

Cheers for your comments,

Nigel.

Well, you can always try reformatting and see if the SMART errors stop. If you keep getting smart errors though, I think an return would be advised.

40-50 MB/s for individual transfers to XP is about what you should get via SAMBA, though with some planned improvements that may get pushed up some more.

The good news is that with lunix you can move from PCI to PCIE based controllers and PMP's and such and keep the same array of disks and not have to reformat, so it's nice that you can get something going cheaply and easily at the start, and migrate to new hardware as budget and needs allows.

[Article22]

I think I will try to reformat one drive, after grabbing the Samsung Utilities disk and doing a low level format on it and see if SMART is any happier.

Been looking into the Intel chipsets and I am now uncertain as to whether my system uses PCI or PCI-E with the disk array.

It looks like the SATA controller is directly integrated into the main bus and so just has a direct 150 Mb/s (SATA I) connection to each drive.

If this is the case then I should be fine with using the on-board connectors for the moment as it supports all of the features I need.

There were some posts on making sure you set each drive to SATA I, is this still the case as I think none of mine are set?

Currently I am using the on-board gigabit ethernet and a PCI Intel Pro Gigabit card using port bonding set to balance-tlb which would probably saturate the PCI bus when more than 1 clients are connected (I have 3 currently)?

Also on the PCI bus are two tuner cards (DVB-S and PAL) so I am slightly worried that there might be some contention going on there perhaps?

Basically I am trying to see where the bottlenecks are and if possible minimise them as cheaply as possible and hopefully others on this forum can benefit from looking at how they can improve their home built NAS's...

Thanks,

Nigel.

[mikesm]

Originally Posted by Article22

I think I will try to reformat one drive, after grabbing the Samsung Utilities disk and doing a low level format on it and see if SMART is any happier.

Been looking into the Intel chipsets and I am now uncertain as to whether my system uses PCI or PCI-E with the disk array.

It looks like the SATA controller is directly integrated into the main bus and so just has a direct 150 Mb/s (SATA I) connection to each drive.

If this is the case then I should be fine with using the on-board connectors for the moment as it supports all of the features I need.

There were some posts on making sure you set each drive to SATA I, is this still the case as I think none of mine are set?

Currently I am using the on-board gigabit ethernet and a PCI Intel Pro Gigabit card using port bonding set to balance-tlb which would probably saturate the PCI bus when more than 1 clients are connected (I have 3 currently)?

Also on the PCI bus are two tuner cards (DVB-S and PAL) so I am slightly worried that there might be some contention going on there perhaps?

Basically I am trying to see where the bottlenecks are and if possible minimise them as cheaply as possible and hopefully others on this forum can benefit from looking at how they can improve their home built NAS's...

Thanks,

Nigel.

First off, what exact hardware are you using? Motherboard, CPU, etc?

I don't think bonding helps unless you are running them in 100 mbps mode, a single gigabit nic should be able to get close to saturation if it's a decent controller. Most of the Intel boards out there with GigE are PCI-E connected, but once you tell me the hardware I can check it for you.

I would keep everything set for SATA II on the disks, at least far as linux is concerned. The motherboard ports are also generally PCIE connected, so they are better to use than an addin PCI card, though some onboard ports have "quirks". If you have an ICH8R or ICH9R, then these have excellent AHCI support in linux and I would recommend using them instead of a PCI board.

The tuners shouldn't matter unless you have serious issues with bus contention, so no worries there.

[Article22]

My server specification is as follows :

Foxconn 915P7AD (Intel ICH6R Southbridge and Broadcom 5788 Ethernet) m/b
Intel Pentium 4 3.0 Ghz HT cpu
4 Samsung 400 GB SATA II HDs
2 x 512 Mb DDR 3200 memory
PCI-E Nvidia 7200 graphics card
PVR-150 PAL tuner card
TT S-1500 DVB-S tuner card
PCI Intel PRO 1000 Gigabit ethernet card

The trouble is that I want to be able to use the network for a multitude of uses so thought it would be best to have the main file server to be able to serve more than one client at the same time using two bonded gigabit cards.

I currently have 5 clients which use the file server...

Any thoughts on the best way forward? Almost tempted to buy a new 775 m/b with some cheap DDR2 memory and reuse the rest of the kit bar the memory.

Nigel.

[mikesm]

Originally Posted by Article22

My server specification is as follows :

Foxconn 915P7AD (Intel ICH6R Southbridge and Broadcom 5788 Ethernet) m/b
Intel Pentium 4 3.0 Ghz HT cpu
4 Samsung 400 GB SATA II HDs
2 x 512 Mb DDR 3200 memory
PCI-E Nvidia 7200 graphics card
PVR-150 PAL tuner card
TT S-1500 DVB-S tuner card
PCI Intel PRO 1000 Gigabit ethernet card

The trouble is that I want to be able to use the network for a multitude of uses so thought it would be best to have the main file server to be able to serve more than one client at the same time using two bonded gigabit cards.

I currently have 5 clients which use the file server...

Any thoughts on the best way forward? Almost tempted to buy a new 775 m/b with some cheap DDR2 memory and reuse the rest of the kit bar the memory.

Nigel.

The 915 is a little long in tooth, and considering the prices for DDR2 ram are so much cheaper, I would definitely build a new system, sell the DDR RAM on ebay or craigslist for $50 or so, and go from there. Get a new Motherboard with PCIE GbE builtin, and an ICH9 SB, and I think you'll have a lot better luck.