Network Transfer Stutter >50% -- Disk Bottleneck?

C

cre3d

Limp Gawd
Joined
Aug 25, 2004
Messages
487
Win7 to Windows Home Server 2011, both clean installs.

From:
Win7 Ultimate x64
WD640x2 Raid0 (black 7200's on ICH10R)
Onboard Intel 82579v Gigabit Ethernet

To:
Windows Home Server 2011
2x Seagate 1.5tb 7200rpm drives on a 3ware 9650se 4lpml in JBOD
Intel Gigabit CT (PCIe)



Transfer starts off great, then turns into a sloppy up down mess like so:



Tried copying back from the server and got this:



Safe to say I'm bottlenecking due to the drive on the windows server? Is there any way I can throttle the transfer speed so as not to hit such a nasty wall? Pretty sure the drive could keep up @ 80mb/s, but 100+ seems to be killing it after ~3gb are transferred. I will be moving ~3tb of data over once my new drives come in (hitachi 5k3000's, slower even) and I'd rather be moving this stuff faster than slower.
 
What kind of data is being transferred?
If its all huge single files, you'll max out your link nicely as shown in the beginning of the graph, if its tons of tiny files, then you'll be thrashing your heads around a lot with seeks/randoms, thus slowing the transfer.
Could be that first part of your transfer is one huge, sequential file.
 
Single 6.8gb file in this scenario. Happens with multiple ~700mb files as well. Jperf test of the same file showed no spiky behavior on the graph, a constant ~110mb/s transfer for the entire duration using a 64k TCP window.
 
Strange, I consitently hit 97%/99% usage in an almost perfect, flat, smooth line when doing large sequential transfers between computers. I've got a server/nas box with 4 F4 2TB in R10 back and forth to two other boxes both with R0 SSDs. Only when I hit sections with tens of thousands of small files (pics, words docs etc.) does the graph start to get jittery and throughput slows (disk bound).

What router/switch?
 
Netgear 8 port gigabit (the blue metal one), also tried it with just a direct cat6 cable from box to box which tested good using intel's cable diagnostics. No change. I plan to do some more testing when I get home today, gonna throw a fresh install of windows 7 on the windows home server box to see if it's possibly an OS issue. If that's not it I plan to Raid0 the 1.5tb seagate drives on the 9650se controller to see if I can rule out the disk bottleneck. I feel like the JPerf results already kinda pinpointed it as a disk issue but I'm not convinced, and even so, the resulting behavior is a bit odd... Shouldn't the OS throttle the transfer speed down to where the disk isn't experiencing queuing issues anymore instead of continuing to push for full throughput and causing the spikes to continue?
 
RAID 1 the drives. I'm wondering if the 3ware card is doing something strange in JBOD mode.

Usually graphs with a big delta between two plateaus mean something was getting cached/ buffered and ran out of space on the faster tier of storage or data being transferred swapped to either the inner part of a disk (or RAID 1 disks)/ started hitting random disks sequentially slowing down speed.

Since your strange graph is writing to the WD black drives, start on that machine. Unless you are doing some odd JBOD access pattern on the server, the issue is on the desktop.
 
It's reading from the WD blacks writing to the Seagate 1.5's. I brought up the performance monitor on the windows home server 2011 machine to watch the Disk Queue Time graph during the transfer, the queue time was 0-1 for the first half of the file and as soon as I saw the spikes begin on the netgraph the queue times were going from 0-3.

Edit: To rule out the 3ware card I tried a transfer that copied to the 2nd partition of the wd black system drive which is on the motherboard intel controller, no change.
 
That looks like hat i have in certain situations:


(click for full screen)

I only have this when i start copying from folder1 to folder2 on my NAS both with Foobar2000 (its music files that need to go from the incoming folder to the artist folder) as with windows transfer.

I honestly have NO clue on what the problem is. It "looks" similar but its not the same. If i find anything close to a hint I'll let you know!
 
After some more testing it seems the 3ware controller was in fact at fault. With disks exported as JBOD the transfer is always spiky and disk queue times get as high as 3-4. After reconfiguring the drives as raid0, transfers were flawless at 105mb/s avg using a 16gb test file. Drives configured as single, again, 105mb/s. Drives then swapped to the motherboard intel ports, again, 105mb/s. Guess JBOD is a nono with this controller and windows home server 2011 :confused:
 
You must log in or register to reply here.
Back
Top