Finding NFS Bottlenecks

[moyam01]

Finding NFS Bottlenecks

Setup:

Server:

Sheevaplug (ARM based mini-computer w/ 1.2 GHz Kirkwood processor) http://www.plugcomputer.org/development-kits/plug-computer-basic.html
Hitachi 3 TB 5400 RPM drives in Silverstone SB 2.0 enclosure (Two- one as active drive and one for backup)
Mbit switch on router connecting to clients
Debian 6.0.1

Client:

Core2Duo E5500
2 GB DDR2 RAM
Gigabit NIC
Kubuntu 11.04

I am peaking at about 24 MB/s according to the KDE notifications while copying files, however I fear that number may not be a true representative of the actual transfer speeds on the setup as those numbers constantly fluctuate between the 24 MB/s and 0 MB/s.

EDIT: Measured time it takes to transfer file with stopwatch, got about 8.80 MB/s.

I would like to know where the bottleneck is on my system. I suspect my Mbit switch would be the most constrictive, but would like the opinion of someone else as I do not have much experience in system optamization of this sort. I want to spend time/money where it would yeild the best result. I figured upgrading to a gigabit switch would ield me the bes result but if tweaking stuff like TCP/UDP or nfs settings would do me better, please let me know

I was thinking of getting something like:

Netgear GS108T-NAS
http://www.amazon.com/Netgear-GS108...8VSK/ref=sr_1_1?ie=UTF8&qid=1322889649&sr=8-1

or

Cisco SG-300-10
http://www.amazon.com/Cisco-SG-300-...RN6U/ref=sr_1_2?ie=UTF8&qid=1322889649&sr=8-2

(I plan to continue o play with networking in the future, thats why im looking at a managed switch)

Any help would be greatly appreciated.

 

[Red Falcon]

The CPU in that Plug system is very weak and I'm honestly surprised it can even push that transfer rate.
This would explain the large transfer fluctuation, especially if the NFS client is setup in SYNC mode instead of ASYNC.

Honestly, your 10/100 switch will be the biggest bottleneck, being that the max transfer rate is 12.5MB/s, and more realistically at 11MB/s.
I think another limitation is the USB 2.0 HDD enclosure. 24MB/s is average for such an enclosure, with about 30MB/s being the top speed that any single USB 2.0 device is capable of.

Even though your system is showing 24MB/s, that's just the file system not understanding SYNC mode in NFS.
The same thing happens to me in SYNC mode, but a more accurate reading is given in ASYNC mode.

 

[moose517]

if its on a 100mb/s switch that would make sense that your average with the stopwatch was about 9MB/s. Your probably peaking out the network connection rather than HDD

 

[Red Falcon]

if its on a 100mb/s switch that would make sense that your average with the stopwatch was about 9MB/s. Your probably peaking out the network connection rather than HDD

Even with gigabit, it will be no faster than the USB 2.0 enclosure will allow, which is about 30MB/s tops, unless eSATA is used.

 

[moyam01]

The CPU in that Plug system is very weak and I'm honestly surprised it can even push that transfer rate.
This would explain the large transfer fluctuation, especially if the NFS client is setup in SYNC mode instead of ASYNC.

Honestly, your 10/100 switch will be the biggest bottleneck, being that the max transfer rate is 12.5MB/s, and more realistically at 11MB/s.
I think another limitation is the USB 2.0 HDD enclosure. 24MB/s is average for such an enclosure, with about 30MB/s being the top speed that any single USB 2.0 device is capable of.

Even though your system is showing 24MB/s, that's just the file system not understanding SYNC mode in NFS.
The same thing happens to me in SYNC mode, but a more accurate reading is given in ASYNC mode.

Thanks for the reply. I was looking at my top output on the server and it seems that my CPU idle percentage stays between 30-50% during transfers, so I am led to believe processor speed is not a constraint, at least not ye. However, is this misleading?

I chose to use SYNC in my setings because it is the safest way to operate betweenthe two machines. Later I will pla around with the ransfer settings if I do not get the desired speed.

 

[danswartz]

No one is saying you have to use async mode, but disabling sync to test at least will answer the question...