D-Link Forums
The Graveyard - Products No Longer Supported => D-Link Storage => DNS-323 => Topic started by: Tank_Killer on August 31, 2008, 08:20:50 AM
-
I just got a 323 and i want to put 2x 1TB in raid 0. I have read the HD compatability chart and see the other 1TB hds on there. Im just wondering if anyone has sucsessfully used a Samsung HD103UJ 1TB HD in their 323?
Thank you for your time
Tank
-
Not the Samsung, but I just built a second DNS-323 with a pair of Seagate 1TB drives (model ST31000340AS)
They are working fine as stripped RAID.
-
By striped RAID I assume you mean RAID0 - for a total of 2GB of space.
I take it you're aware that failure of either disk will result in the loss of all your data?
-
i am aware of the risk when striping disk's. I have other forms of backup and no crucial data will be stored on the unit.
thanks tho.
I just bought a coupla of seagate 1tb, just so there is no screwing around.
-
i am aware of the risk when striping disk's. I have other forms of backup and no crucial data will be stored on the unit.
thanks tho.
I just bought a coupla of seagate 1tb, just so there is no screwing around.
OK - you're aware of the risk - is there any advantage to striping it?
-
OK - you're aware of the risk - is there any advantage to striping it?
Yes. If you want to store a 2TB file. ;D
-
Yes. If you want to store a 2TB file. ;D
You could do that with JBOD - I'm looking for advantages specific to "striped RAID"
-
You could do that with JBOD - I'm looking for advantages specific to "striped RAID"
JBOD, in terms of risk, is roughly equivalent to RAID0 on a DNS-323. ;D
-
JBOD, in terms of risk, is roughly equivalent to RAID0 on a DNS-323. ;D
I don't know if you noticed, but my question is not related to risk - not that I'm disagreeing with you in any way.
-
You may notice a DECREASE in performance with Raid-0. YMMV based on your IO patterns. I did some benchmarks of my own a while back on all the available FW versions and found worse results in going with RAID-0 compared to RAID-1 and Individual disks. Best performance was single disks. Next best was RAID-1. EXT2 is faster than EXT3.
Cheers!
BV
<massive edit job. my first attempt was from memory and I had things mixed up. Found my test result sheet and fixed the error>
-
You may notice a DECREASE in performance with Raid-0. YMMV based on your IO patterns. I did some benchmarks of my own a while back on all the available FW versions and found worse results in going with RAID-0 compared to RAID-1 and Individual disks. Best performance was single disks. Next best was RAID-1.
Ya, that tends to happen when reading a bunch of small files due to seek times. However, you should be getting close to double the throughput with RAID-0. This is very usefull when reading/writing large files (say 20MB and above).
If this turns out to be not the case with the DNS-323, then this is an issue I think needs to be looked into further. Can anyone else confirm or deny these findings?
-
Ya, that may be true for other devices, but not on the DNS-323. I tested with a mix of file types and sizes and raid-0 was consistantly WORSE than the other configs...
-
This is actually what I was trying to get out of Tank_Killer
Like bspvette86 my tests have shown no performance benefit to RAID0.
RAID0 will only provide improved performance when the bottleneck is the DISK DRIVE - it does this by allowing reads/writes alternately from/to each disk, so that rather than the data transfer being interrupted for the disk to step the heads to the next track, etc, this "housekeeping" is done whilst the data is being transferred using the other disk.
Once the disk is faster than the electronics - which in this case also include the network - RAID0 no longer provides enhanced perfomance.
-
I striped the discs and tried single disc configuration (tests were done gbic to gbic). With 9k jumbo frames (and flow control disabled on all non gbic clients) I noticed negligible performace difference when writing large files (4+ gb), raid0 came out slightly on top for me. now i have a high quality switch and cabling, that might make a diffference (I can also assign weighted packet priority for everything coming from and going to the DNS-323). If i had control over the sector size of the raid I might be able to further increase that gap. Im guessing performace in the enviroments with this sort of network appliance are gunna vary greatly depending on network topography ect..
what a neat convorsation started by just asking about disk compatability :D
Tank
-
Tank,
Change all you want until you are blue in the face but it will "NOT" make much difference. You can even do a direct Giga-giga connection between your CPU and the DNS and it will not make a difference. The DNS-323 is CPU/firmware bound.
Regards,
BSPvette
-
I'm using my DNS-323 with Samsung 1TB HD103UJ x2.
It would hang on format at 94% with out of box firmware (1.04.) Updating to 1.05 resolved this issue.
As for raiding them, it does work, however, I chose to have them as separate disks.
-
I'm using my DNS-323 with Samsung 1TB HD103UJ x2.
It would hang on format at 94% with out of box firmware (1.04.) Updating to 1.05 resolved this issue.
As for raiding them, it does work, however, I chose to have them as separate disks.
I'm running into the same problem with my DNS-323 with Samsung 1TB HD103UJ x2, that is, it would hang on format at 94% using out of box firmware (1.04). Updating to 1.05 does resolved this issue. I'm running it in raid1 mode...so far so good. The only thing that I've noticed is that I'm only getting around 12-15MB/s throughput with my DIR-825 gigabit router (laptop network card is a Marvell gigabit client) - jumbo frames disabled.
-
The only thing that I've noticed is that I'm only getting around 12-15MB/s throughput with my DIR-825 gigabit router (laptop network card is a Marvell gigabit client) - jumbo frames disabled.
What's the typical file size? 12-15MB/s is in the ballpark, depending on file size...
-
Yeah...after reading the DNS-323 specs more closely, 12-15MB/s seems to be the norm for writes. The throughput for reads are a bit faster (up to 23MB/s). I guessed the bottleneck is within the DNS-323.