Latest interface: 0.3.1
Latest system: 010
bronan
User

3 posts

Posted on 11 July 2015 @ 16:00
I have run for many years 2 synology nas both 4 disk series.
Recently i detected that some of the stored data turned out to have lost data in them ... and after reading alot of inof about it know raid5 is not secure as it was, with the massive growth of the disk sizes.
My current DS411 and DS414 have 4 x 4 TB leaving 10.8 Gb available and i found 2 corrupted files.
However there is too much data on them and weird enough both nas report being healthy.
The first corrupted file i found was an image i had made from my main workhorse which has 3 x 4 Tb data and 2 vertex4 250 gb boot drives in raid0
The second corrupted file which i detected which i lost was a BD50 copy of a movie i had from bdrom because the bdrom died i thought i could watch the movie by loading it into a virtual rom.
But as i said its corrupted as well the file appears to be correct but its not readable anymore.
I have seen many examples of self build nas based on supermicro materials.
So i would like to start with a second hand base system server and seen alot of dell, hp and supermicro based 2u servers offered, but i am not sure if those dual cpu machines are a good choice.
Because i read most having problems with all kinda things likes drivers, usb, controllers the list is long.

However i also would like to use larger drives like the new HGST 10 Tb disks.

Also would like to make it large enough to have enough space for coming years so 8 to 12 disks
I am not sure if it would be wise to start with 6 and later add another 6 in a second pool or that its better to have one larger pool

when i look at the offers some really look hardware wise pretty good as a base for a good stable nas like for instance : HP server Proliant DL380 G6 2x X5650 2,66GHz / 72GB RAM

What do you think guys btw i am dutch so excuse my crappy english :)
CiPHER
Developer

1199 posts

Posted on 12 July 2015 @ 00:16
Legacy storage (2nd generation filesystems) are not designed to handle with corruption. All they can do is make sure the metadata returns to a consistent state after recent writes have been lost. This is common with harddrives that lose the data in DRAM buffercache. But these filesystems where never designed to offer any real protection to your data. They are very outdated and simply don't cut it anymore in the realm of high capacity harddrives where the BER (Bit-Error Rate) is high enough to cause problems. RAID5 is inherently insecure due to the RAID5 write hole.

Synology uses ext4+md-raid+SHR which is just legacy storage. If your data corrupts, the filesystem will not even know about it. It only knows if the metadata is consistent. The RAID layer only knows whether the data matches the parity data (RAID5/6) but it doesn't know which of the two is the correct version. It just assumes the parity is incorrect and updates the parity to match the stored data; thus potentially causing the corruption and overwriting the one only good version.

Long story short, if you want to give your files some protection, you really need a 3rd generation filesystem like ZFS, Btrfs, ReFS, Hammerfs. Only ZFS is mature and really usable; the best way to store your data in this era.

However, the hardware you have in mind is kind of excessive. Why do you need such a heavy machine with so many cores and so much RAM?

In fact, your system can run on a modern Atom-like chip such as the Intel J1900 (10 watts) quite well. Now i do recommend ECC for a build like this, but i'm sure if you buy things new you can keep the price very low. And it will consume much less power than older second hand server hardware which usually is not power efficient at all.

So maybe look for a nice single processor (UP) board from SuperMicro?

Do not use your OCZ SSDs for anything other than less important things, they cannot be trusted! They are unprotected SSDs with complex firmware that is prone to firmware bugs. They can corrupt easily. Do not use those SSDs as boot device for your NAS! You can use them as L2ARC however, since ZFS will detect corruption from SSDs used as L2ARC. It will read from the pool instead.

And harddrives; those 10TB drives are SMR drives and i recommend that you avoid them as long as possible. Go for the PMR drives instead, the WD Green 6TB is a cheap and good drive with very modern 1200GB platters and 175MB/s peak performance at the beginning of its capacity, which is very good for a 5400rpm-class drive. The lower the rpm, the better for mass storage. Your performance will come from the higher data density (more data per revolution) and from the number of harddrives that are being utilised in parallel. Meaning: ZFS uses multiple drives in RAID-Z which will amplify the sequential speeds considerably.

If you store primarily large files, then the story above applies to you. If you also store things like VM images, then i recommend you buy a (couple of) nice SSDs like Intel 320, Intel S3500 or Crucial MX100 or the MX200 128GB version. Those are protected SSDs very suitable for ZFS. You can create a separate pool for stuff like VMs etc, leaving only mass storage on your big HDD pool.

You can create multiple RAID-Z2 or have one big RAID-Z3 with 19 drives. The latter would provide the highest capacity relative to the level of protection. In other words: it is space efficient. Only 3 drives parity for 16 drives data. 16 drives also means you have a power-of-two which is important for how ZFS stores data. If you use 6TB drives, that means 16*6 = 96TB usable space - though remember you should never fill your pool to the limit. So let's say 80TB of optimal data storage. That is quite future proof i think, but that depends on your personal needs of course.

Let me know what other avenues you would like to explore. Good luck with your ZFS project! :)
bronan
User

3 posts

Posted on 13 July 2015 @ 02:07
Well i kinda knew about the problem however never expected it to be failing so often, somehow i am sure many other images will turn out to be corrupt.
But time will tell, why the choice of these old servers well ... they are kinda cheap ;)
They are for sale for between 300 to 1000 euro and have a nice huge bunch of memory for that low price. Ofcourse most of them have only a few small drives. But for me thats ok the os does not need alot of space.

I agree they are a bit overkill but funny enough the ones with cheaper and slower xeons are more expenssive. A machine with a single xeon 1.6 ghz is often around 600 to 1200 and only have 8 or 16 Gb mem. While that other offered machine has more than enough for less.
Even though it uses a bit more power i am sure won't matter much, my 5k solar array will compensate some of it anyway ;).
I was indeed asking myself would i split the storage into different vpools.
The question which is kinda hard to answer is how many vm storage i need, because that changes pretty often.
Ofcourse some changes will allways happen on the data but i wonder if its wise to use some ssd'd as cache for the large drive pool.
Several people have been using my storage space as backup for alot of things, and some people do make frequent system backups on them. They use all kinda backup solutions apple/windows/linux/unix. So its basically storage for longer term in most cases. Some of them use a backup scheme which deletes after set periods some of the made backup files.
On the current nas they are stored in a recycle bin by me, because i had situations that some of the people by mistake had deleted important data.
But in most cases i clear the recycle bin files after 6 months.

Will follow your advise on the wd greens ofcourse i have still alot of time to complete the large drive project.
But since i know the current machines are not safe. I kinda want to quickly switch over to a safer spot for everyone involved.
Kinda fear that one of them might loose important data or already lost info without us knowing. The raid5 flaw is too risky to keep going like we have done now. Because my used drives which i used in the current systems are well over a year old or older i am planning on selling them as soon as i got a replacement.
Here are current the wd green 6tb are about 240 euro each at the moment. So for the 19 drive nas i need to safe up some more money. However for the project in mind for that i still have some time.

If i would go for a modern machine hashwell based.

It would be a SuperMicro X10SRL-F which is 340 euro with Crucial 32GB DDR4 ECC Registered kit which costs me 330 and a xeon e5-2603 which is 240 euro.
Ofcourse i based it on new because second hand supermicro board are hardly sold in my country and if they do they ask pretty steep prices.
I have no clue if it really supports 10 drives or that this is just on paper.

Giving the fact that my current msi board had in big letters on the box up to 32 Gb ram but in reality even with only 16 Gb ram turned out to be a disaster to get it running.

Might need to get a system running capable of storing what is now on the running systems and what i have stored on disconnected drives and move that onto a system based on Raid-Z till i have enough funds to build the 19 drive system. I am assuming some of the current stored data might already be corrupt., so the sooner the better to make the switch.

Again in the netherlands there is not much for sale at this besides the servers being sold by for homeserver.nl most of those are dell systems i noticed
downloadski
User

17 posts

Posted on 15 July 2015 @ 03:59edited 16 July 2015 @ 04:24
CiPHER
Developer

1199 posts

Posted on 15 July 2015 @ 09:55
Removed a special character in your post - this board has issues with that:

downloadski wrote: There is a supermicro dual xeon e5-2620 for sale on tweakers.net
That is 699 with 2 cpu's
Last Page

Valid XHTML 1.1