Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[crishna]

my best guest, would be that you have 2 drives, in stripe 0 .... i have 2 drive, in stripe 1 (mirror)...

[47th_ronin]

hmm, samba daemon on dns-323 shows up to 14% cpu usage during file copy and menumeters shows transfer rate 6-7 megabytes/s. cables are ok, switch elimination will be next step..

[fordem]

my best guest, would be that you have 2 drives, in stripe 0 .... i have 2 drive, in stripe 1 (mirror)...

No - if I recall correctly that was a RAID1 array.

Think carefully about this - if - as you surmise, the CPU is the bottleneck - would RAID0 provide any performance benefits?  NO - RAID0 writes the data to each disk alternately and will provide enhanced performance ONLY if the disks are the bottleneck.

You can run your own tests, but when last I checked, there was no benefit to RAID0 on the DNS-323.

[47th_ronin]

another note: copying file FROM osx 10.6 imac TO dns-323 (two WD2500AAKS-00L9A, raid 0, fw 1.08, funplug, transmission, upnp) resulted in average speed 20 MB/s which is supposedly as good as it gets in 'normal' home environment. all the components are the same, just file transfer direction is reversed..

[fordem]

Hang on - copying TO the NAS gives you 20MB/s, copying FROM the NAS gives you 6~7MB/s, same everything except direction???

That's pointing to a write bottleneck of some sort on the destination imac

[eclipsed7]

I get the same, write speeds are acceptable at 14-16MB/sec, but my read speeds can be anywhere from 6 to 20 MB/sec, depending on the day.

[urbang33k]

I'm getting 34MB/s read from the DNS off a non-raided Seagate 1TB 7200 rpm 32mb cache drive.

The DNS is set to jumbo frames at 9000

It's connected to a Netgear 8 port gigabit green switch (QoS but is unmanaged). The switch says right on the box that it supports jumbo frame upto 9000

My motherboard is an older ASUS P5W DH which has that original intel 975x chipset with one of the original dual-core chips. This motherboard sports dual ethernet interfaces.  They are Marvell Yukon 88E8053's sitting on the PCI-E bus (supposedly).  These controllers do support jumbo frames and are set at 9000 to match the switch and the DNS323. They also have flow control, offload and whole host of other configuration options that I have no idea what they do and haven't messed around with too much.

once past my interface, the data is flowing to 2 x SATA-150 WD caviar blacks in RAID-0.  Not sure what the specs are but I dont think they are anythign special.   My single WD raptor felt as fast as these two blacks in raid-0.

I too originally experienced slow speeds and I set out to fix it. during my trouble shooting, I found I noticed the largest increase in performance when I enabled jumbo frames at 9000 on BOTH dns and ethernet interface. I'd like to note that my router upstream from my switch is also gigabit and each of its 4 LAN ports can be set to jumbo frames at independant rates.  It's incredibly important to ensure that when selecting jumbo frames, you set each device at the highest common rate.  so, if you have a router that supports jumoframes at 7000 and the dns supports 9000 frames and your pc interface supports 9000, I THINK you should really be setting all devices at 7000.   I THINK.   Only when I set all my devices at the common rate, did my speeds become consistant. YMMV

[urbang33k]

Just to add further... 

I just added my second seagate baracuda 7200 32mb cache  to the dns 323 in a raid 0 config.  I am now getting 20-21MB/s sustained writing and 35-36 sustained reads, files over 2 gig.

so are these the upper limit to what this device is really capable of? I'm going to wipe out the array and build a raid 1 to see how much I lose before loading all my data back on.

Also going to make some ram disks and compare transfers over the network form pc to pc on ramdisks/harddrives for comparison.

[fordem]

urbang33 - based on similar tests done a long time ago - I'd say you're pushing the limits of the silicon.

From memory, the network interface on the DNS-323 can handle perhaps 400 mbit/sec (NIC tests have to be done - so to speak - RAM-to-RAM to avoid disk interface bottlenecks), and the disk interface is slightly slower at around 40MByte/sec (read to nul).

I'll be watching to see how your RAID1 tests compare to RAID0 - as I recall there was no improvement whatsoever.

[urbang33k]

I'll be watching to see how your RAID1 tests compare to RAID0 - as I recall there was no improvement whatsoever.

rebuilt as raid 1 this morning. 

sustained writes to the dns at 18-19 MB per second
sustained reads from the dns at 32 - 33 MB per second.

For me the choice is simple.  Losing hardware failover redundancy is just not worth the extra 1-3 MB/s (up or down)

Ill do some ram drive copying and post some stats along with my conclusion after my breakfast

[urbang33k]

So heres a breif description of the test environment. 

HTPC -  gigabit PCI-E ethernet (jumbo frames at 9016)- amd dual core - 4 gig of memory
Windows 7 64-bit
physical drive - wd raptor 150
virtual drive - 1.5 gig ram drive

Desktop PC - gigabit PCI-E ethernet (jumbo frames at 9016)- intel dual core - 4 gig of memory
Windows 7 32-bit
physical drive - RAID-0 array comprised of 2 x WD Caviar Blacks 500gig each
virtual drive - 1.5 gig ram drive

NAS - DNS323
drive configuration - RAID 1 with 2 x Seagate baracuda 7200 1 TB 32mb cache
Gigabit set at 9000
LLTD - Disabled
UPnP - Disabled

Netgear GS608 V3 switch (Supports QoS and is unmanaged)
Jumbo frames at 9016


results in next post..........

[urbang33k]

First of all, Ill state that reading/writing fronm the DNS323 to my Desktops PC's ramdrive made no difference in transfer speeds as compared to reading and writing from my RAID-0 array in my desktop PC in my previous post.

Heres where it starts to get crazy and depressing for me:

Speeds as reported by windows explorer between ram drives of HTPC and Desktop PC with windows file and printer sharing
DESKTOP -> HTPC  --- Burst  190MB/s  --- Sustained  113-122MB/s
HTPC -> DESKTOP --- Burst 204MB/s  --- Sustained 120 - 133MB/s

Not sure whats going on here as I think these numbers are above the theoretical limits of gigabit. Unfortunately I lack a large amount of ram so I can't get a better reading of sustained transfer rates by using a larger file than 1.5 gigabytes. HOWEVER one thing is certain,  my network switch, CPU, interfaces, memory are definitely NOT the bottlenecks when it comes to the slow transfer rates to the DNS323.

speeds between physical drives of the two computers in next post using larger files....

[urbang33k]

OK so heres some more down to earth numbers
Tranfersfer an 8.5 gig MKV file

DESKTOP -> HTPC --- Burst 118MB/s -- Sustained 80 MB/s
HTPC -> DESKTOP ---- Burst 133MB/s -- Sustained 106MB/s

I wish I had some SSD's to do further testing with but this definitely confirms that that little DNS-323 is the bottleneck here with the 31MB/s reads and 19 MB/s writes.

I will say that when I bought the DNS323 I had every expectation of gigabit speeds simply because it states gigabit on the box.  I had no clue or experience in setting of network shares, and DID NOT do my research ahead of time so I have no one to blame but myself.

The fact that the DNS323 does faster than 10/100 speeds and therefore can say it does gigabit speeds is a pill that a little hard for me to swallow to be honest.  After doing these quick tests I already know my next course of action...  which is to sell this unit and build myself a small raid 5 file server.

In short, I guess the best we can hope for is the numbers I've posted.  Anyone know a way to overclock this rig??  =P

/urbang33k

[fordem]

Dont forget to post the read/write speeds you measure with your RAID5 array - somehow I think you'll be in for a big surprise.

[McPillager]

A bit off topic but...

in your screenshot below I can see two adapters listed which probably means you have two network cards. Just out of curiosity and for the sake of speed measurements comparativeness, what is your network setup?

well, i've been having the same issue... using gigabit ethernet, pluged in directly to my computer, static ip both sides, and the xfer speed is like 10~14 mb/s .... BUT as i read your stuff, i though of something, and indeed, the NAS CPU is peaking at nearly 100% usage on the samba process... there is our bottleneck.