GoFlex Net - slow file transfers

A few months ago I stumbled on the doozan forum researching other things related to linux, and found the idea of using these kirkwood devices as a file server very intriguing.

So I recently purchased a new-in-box GoFlex Net (because it has 2 SATA ports) and followed the instructions to set up the serial connection, install the newest uboot, and the 6.5.7 rootfs. (although I admit my environments may not be perfect because it will not boot with the uEnv.txt file)

There seems to be something going on with the file transfers, as they seem quite slow.

I have my current set up as follows:

- uboot.2017.07-tld-1.goflexnet
- Debian-6.5.7-kirkwood-tld-1-rootfs

-load_dtb_addr=0x2c00000

- booting to a 4gb USB thumbdrive
- dphys-swapfile installed with the swapfactor set to 5 (thinking of using HDD for swap later)

- have also installed vim, ntfs-3g, samba, fake-hwclock, and transmission-daemon

- used bind to load log files to /tmp/var/log (and added commands to /etc/rc.local at the end)

- set the file at /etc/sysctl.d/local.conf with the following
vm.min_free_kbytes = 16384
vm.swappiness = 10

net.ipv6.conf.all.disable_ipv6 = 1
net.ipv6.conf.default.disable_ipv6 = 1
net.ipv6.conf.lo.disable_ipv6 = 1

-fstab entry for the hard drive has - ntfs-3g noatime,big_writes,permissions 0 1

- zswap enabled

- the samba config file has very few lines

#samba config file

[global]
	workgroup = WORKGROUP
	server string = Samba Server %v
	netbios name = Goflex
#	netbios aliases = Fileserver
	security = user
	map to guest = bad user
	dns proxy = no
#	ntlm auth = yes
	Hosts allow = 127., 192.168.1.


[6TB-A]
 path = /mnt/drives/6TB-A
    browseable = yes
	writeable       = yes
	guest ok        = no
	read only = no

	
[6TB-B]
 path = /mnt/drives/6TB-B
    browseable = yes
	writeable       = yes
	guest ok        = no
	read only = no

I have checked both the network throughput and the drive performance, and they both seem fine.
While booting it shows the ethernet adapter with flow control disabled.
TSO is on.


Only 1 drive in the Goflex Net currently - 6TB Toshiba mounted at /mnt/drives/6TB-A (I plan to have another 6TB drive in the future)

hdparm shows the following:

hdparm -tT /dev/sda

/dev/sda:
 Timing cached reads:   472 MB in  2.00 seconds = 235.63 MB/sec
 Timing buffered disk reads: 298 MB in  3.00 seconds =  99.19 MB/sec

I have another linux computer running CentOS 7 sharing a 4tb drive over the network with IP 192.168.1.2. This is currently set up in the smb.config of the GoFlex Net to mount this network drive at /mnt/drives/ on boot.

root@debian:/mnt/drives/6TB-A# time cp /mnt/drives/4tb/bigfile .

real    0m46.279s
user    0m0.000s
sys     0m10.886s

root@debian:/mnt/drives/6TB-A# time cp bigfile /mnt/drives/4tb

real    0m27.544s
user    0m0.011s
sys     0m8.372s

I have also set up a iperf server on the CentOS (gigabit Netgear XR500 router in the middle) and ran the following:

root@debian:/mnt/drives/6TB-A# iperf -c 192.168.1.2 -n 1000M -r
------------------------------------------------------------
Server listening on TCP port 5001
TCP window size:  128 KByte (default)
------------------------------------------------------------
------------------------------------------------------------
Client connecting to 192.168.1.2, TCP port 5001
TCP window size: 43.8 KByte (default)
------------------------------------------------------------
[  1] local 192.168.1.8 port 50382 connected with 192.168.1.2 port 5001 (icwnd/mss/irtt=14/1448/259)
[ ID] Interval       Transfer     Bandwidth
[  1] 0.0000-10.3220 sec  1000 MBytes   813 Mbits/sec
[  2] local 192.168.1.8 port 5001 connected with 192.168.1.2 port 60652
[ ID] Interval       Transfer     Bandwidth
[  2] 0.0000-14.1505 sec  1000 MBytes   593 Mbits/sec

Running the iperf test with the bigfile:

root@debian:/mnt/drives/6TB-A# iperf -c 192.168.1.2 -F bigfile
------------------------------------------------------------
Client connecting to 192.168.1.2, TCP port 5001
TCP window size: 16.0 KByte (default)
------------------------------------------------------------
[  1] local 192.168.1.8 port 41738 connected with 192.168.1.2 port 5001 (icwnd/mss/irtt=14/1448/271)
[ ID] Interval       Transfer     Bandwidth
[  1] 0.0000-20.8813 sec   488 MBytes   196 Mbits/sec

So the issue appears to be with the drive throughput.

I share the drives over the network to several Windows based computers and also Android Smart TVs (used for TV shows and movies).

Current Windows transfer rates (copying from 6TB Goflex to anything on the network) are showing between 14MB/s and 20MB/s. Most 4K movies require higher speeds than this.

Copying to/from the CentOS hosted 4TB drive usually runs at 60-80MB/s.

Any ideas/suggestions to get this higher?

Thanks

Quote

Current Windows transfer rates (copying from 6TB Goflex to anything on the network) are showing between 14MB/s and 20MB/s. Most 4K movies require higher speeds than this.

Quote

Copying to/from the CentOS hosted 4TB drive usually runs at 60-80MB/s.

Is the CentOS box also using SAMBA when you get 60-80MB/s? or using NFS to copy to/from the GFNet?

====

There are several Wiki thread entries:

Quote

Perfornance Tuning & Benchmarks

Network performance - SAMBA - NFS (various protocols)

Samba Tuning
Mount NTFS with big_writes

"Samba Tuning" post is old, but perhaps it will help.

-bodhi
===========================
Forum Wiki
bodhi's corner (buy bodhi a beer)



Edited 1 time(s). Last edit at 03/14/2024 02:48PM by bodhi.

Quote

- dphys-swapfile installed with the swapfactor set to 5 (thinking of using HDD for swap later)

Also, the swap file should be on HDD while running the test. This box 's RAM is limited.

-bodhi
===========================
Forum Wiki
bodhi's corner (buy bodhi a beer)

Hi Bodhi,

Thanks for your quick response. I have been reading many of your posts over the past few weeks and see that you are definitely the #1 guru when it comes to these kirkwod devices!


bodhi Wrote:
-------------------------------------------------------
> Is the CentOS box also using SAMBA when you get
> 60-80MB/s? or using NFS to copy to/from the
> GFNet?

Yes, the CentOS is set up to share the drive on the network using samba. I was hoping to use the GoFlex Net to replace this machine due to power/heat.



I have reviewed the other threads covering performance and tuning. This is where I found the suggestion to put all of the logs (including the samba log) in the /tmp folder. Since I am using the procedure to load log files to /tmp/var/log using the /etc/rc.local file, there is no need to add the command in the smb.conf to put the log file there (I think).

The Goflex drive is already mounted using the 'big_writes' option in fstab.
I can post a copy of my fstab if you need to see it.

I have not tried using NFS instead of Samba - does this make file transfers a little faster?

I will need to create an EXT3 partition in the 6TB drive and see about at least placing the swap file there, maybe even the entire rootfs. Which do you suggest for best performance?

Actually I just edited the /etc/dphys-swapfile to specify the location of the swap file:

# where we want the swapfile to be, this is the default
CONF_SWAPFILE=/mnt/drives/6TB-A/swap

...and made sure that there is a folder there named 'swap'.

Will reboot and run the performance tests again.

Hi Tedious_1,

> > Is the CentOS box also using SAMBA when you get
> > 60-80MB/s? or using NFS to copy to/from the
> > GFNet?
>
> Yes, the CentOS is set up to share the drive on
> the network using samba. I was hoping to use the
> GoFlex Net to replace this machine due to
> power/heat.

For SAMBA, 60-80MB/s is quite decent. Hard to get more than that range.

> I have not tried using NFS instead of Samba - does
> this make file transfers a little faster?

Definitely faster. You can maximize throughput using NFS.

But I'd imagine your HTPC is a Windows box? I never have problem streaming HD videos (1080p) from this GF Net and its cousin GF Home to my Windows HTPC. But I don't have 4K so don't know how it'll behave.

If your HTPC is a Linux box then I'd say use NFS, don't use SAMBA. You can run both NFS and SAMBA on the GF Net, so that you can copy files to/from Windows using SAMBA. But set up the video streaming from NFS shares.

>
> I will need to create an EXT3 partition in the 6TB
> drive and see about at least placing the swap file
> there, maybe even the entire rootfs. Which do you
> suggest for best performance?

The rootfs can be on USB with the precaution to lessen the wear and tear (which I see you already done). The HDDs can be put to sleep (if the system is really idle, the HDD swap file is not accessed by the kernel).

If you want to put the rootfs on HDD, then it cannot go to sleep. So it's the tradeoff you'd make. And use about 20GB for rootfs partition if it's on HDD.

-bodhi
===========================
Forum Wiki
bodhi's corner (buy bodhi a beer)

Hi Bodhi,

Thanks for working with me on this. I'm definitely interested in getting rid of that desktop with CentOS installed and use the GoFlex Net with 2 drives. It would definitely use less power.

Yes, I stream some TV shows from the drive on 1080p to a windows based computer, but the 4k movies go directly to the smart TVs with Kodi (android). I have not benchmarked the file transfer speeds when watching a 4k movie, but I am sure depending on the compression used in the file that a transfer rate of only 19MB/S is not sufficient.

I also use my Windows desktop to manage these files and transferring at 60-80MB/s helps with reducing the time it takes to move these enormous files. Most 4k movies with DTS audio are 15GB or more!

I added the swap file location to a folder on the hard drive and rebooted the box, then ran the tests again.

root@debian:/mnt/drives/6TB-A# time cp /mnt/drives/4tb/bigfile .

real    0m43.359s
user    0m0.012s
sys     0m10.326s
root@debian:/mnt/drives/6TB-A# time cp bigfile /mnt/drives/4tb

real    0m29.948s
user    0m0.000s
sys     0m9.063s
root@debian:/mnt/drives/6TB-A# iperf -c 192.168.1.2 -n 1000M -r
------------------------------------------------------------
Server listening on TCP port 5001
TCP window size:  128 KByte (default)
------------------------------------------------------------
------------------------------------------------------------
Client connecting to 192.168.1.2, TCP port 5001
TCP window size: 43.8 KByte (default)
------------------------------------------------------------
[  1] local 192.168.1.8 port 50442 connected with 192.168.1.2 port 5001 (icwnd/mss/irtt=14/1448/274)
[ ID] Interval       Transfer     Bandwidth
[  1] 0.0000-10.0749 sec  1000 MBytes   833 Mbits/sec
[  2] local 192.168.1.8 port 5001 connected with 192.168.1.2 port 32942
[ ID] Interval       Transfer     Bandwidth
[  2] 0.0000-14.0825 sec  1000 MBytes   596 Mbits/sec
root@debian:/mnt/drives/6TB-A# iperf -c 192.168.1.2 -F bigfile
------------------------------------------------------------
Client connecting to 192.168.1.2, TCP port 5001
TCP window size: 16.0 KByte (default)
------------------------------------------------------------
[  1] local 192.168.1.8 port 49364 connected with 192.168.1.2 port 5001 (icwnd/mss/irtt=14/1448/281)
[ ID] Interval       Transfer     Bandwidth
[  1] 0.0000-22.8524 sec   488 MBytes   179 Mbits/sec

Any idea where this could slowness could be coming from? The ethernet adapter seems to be coming close to maxing out the transfer rates of the gigabit adapters/router, and the hdparm speeds seem decent enough.

I did notice during boot that the SATA adapter is loading the drive at UDMA100 instead of UDMA133, but not sure if that has any bearing on the issue at hand.

[    6.160831][   T55] ata2: SATA link up 3.0 Gbps (SStatus 123 SControl F300)
[    6.320877][   T55] ata2.00: ATA-8: MG04ACA600E, FS3J, max UDMA/100
[    6.327226][   T55] ata2.00: 11721045168 sectors, multi 16: LBA48 NCQ (depth 31/32)
[    6.341784][   T55] ata2.00: configured for UDMA/100

This is an enterprise drive with max 6Gb transfer
https://storage.toshiba.com/docs/enterprise-hdd-documents/ehdd-mg04aca-product-manual.pdf


Any information or suggestions that you (or anyone else on the forum) can give would be greatly appreciated.

Thanks,

>
> There are several
> Wiki
> thread entries:
>
>

Quote

Perfornance Tuning & Benchmarks
>
> Network performance - SAMBA - NFS (various
> protocols)
>
> Samba Tuning
> Mount NTFS with big_writes

>
> "Samba Tuning" post is old, but perhaps it will
> help.

Have you tried to use the smb.conf attached to the "Samba Tuning" post? save your smb.conf file and copy it on top. Change the shares names.

https://forum.doozan.com/read.php?8,23223,23293#msg-23293

Quote

2. Set up a public share (no login credential). This is a wide open share that everybody in the network will have read/write privelege. Here I've attached my simplest smb.conf that you can use. Download it to your box, save the current one, and copy the new smb.conf to this directory

-bodhi
===========================
Forum Wiki
bodhi's corner (buy bodhi a beer)

Thanks Bodhi!

I don't really like having an open share on my network with these movie and tv files. Once had someone connecting on wi-fi that deleted an entire tv series.

But I can do this to test the speeds and see the results.

I copied the smb.conf file to the GoFlex Net and edited the path of the shared folder.

[media]
        comment = shared media (eg. USB drives)
        path = /mnt/drives/6TB-A
        browseable = yes
        writeable = yes
        guest ok = yes

After restarting I am able to see a shared folder named 'media' on all network machines.

Running tests again gives the following:

root@debian:/mnt/drives/6TB-A# hdparm -tT /dev/sda

/dev/sda:
 Timing cached reads:   514 MB in  2.00 seconds = 257.07 MB/sec
 Timing buffered disk reads: 296 MB in  3.02 seconds =  98.17 MB/sec
root@debian:/mnt/drives/6TB-A# time cp bigfile /mnt/drives/4tb

real    0m28.630s
user    0m0.012s
sys     0m8.395s
root@debian:/mnt/drives/6TB-A# time cp /mnt/drives/4tb/bigfile .

real    1m1.690s
user    0m0.009s
sys     0m12.666s
root@debian:/mnt/drives/6TB-A# iperf -c 192.168.1.2 -F bigfile
------------------------------------------------------------
Client connecting to 192.168.1.2, TCP port 5001
TCP window size: 16.0 KByte (default)
------------------------------------------------------------
[  1] local 192.168.1.8 port 56540 connected with 192.168.1.2 port 5001 (icwnd/mss/irtt=14/1448/264)
[ ID] Interval       Transfer     Bandwidth
[  1] 0.0000-21.4232 sec   488 MBytes   191 Mbits/sec

Thoughts?

Tedious_1,

> root@debian:/mnt/drives/6TB-A# iperf -c
> 192.168.1.2 -F bigfile

I think the big file here might not be a good test for iperf.

This was a good test because you are not using file system:

Quote

I have also set up a iperf server on the CentOS (gigabit Netgear XR500 router in the middle) and ran the following:
root@debian:/mnt/drives/6TB-A# iperf -c 192.168.1.2 -n 1000M -r
------------------------------------------------------------
Server listening on TCP port 5001
TCP window size: 128 KByte (default)
------------------------------------------------------------
------------------------------------------------------------
Client connecting to 192.168.1.2, TCP port 5001
TCP window size: 43.8 KByte (default)
------------------------------------------------------------
[ 1] local 192.168.1.8 port 50382 connected with 192.168.1.2 port 5001 (icwnd/mss/irtt=14/1448/259)
[ ID] Interval Transfer Bandwidth
[ 1] 0.0000-10.3220 sec 1000 MBytes 813 Mbits/sec
[ 2] local 192.168.1.8 port 5001 connected with 192.168.1.2 port 60652
[ ID] Interval Transfer Bandwidth
[ 2] 0.0000-14.1505 sec 1000 MBytes 593 Mbits/sec

So you would want to go to the other box and copy the file from the GoFlex Net to it and see how SAMBA handle the transfer.

-bodhi
===========================
Forum Wiki
bodhi's corner (buy bodhi a beer)

Hi Bodhi,

Thanks for your help so far.

I was searching for other information that could possibly shed light on the issue, and I stumbled across something that Doncharisma says on his own website:

Quote
doncharisma
My GoFlex NET has now been running a couple of weeks and seems stable enough with OpenMediaVault and my two 1TB RAID1 disks. Performance wise I’m happy enough, only bottleneck I’ve found so far is the Samba daemon which is using a lot of CPU when I copy files over the network onto the RAID array. But still I’m getting around 20MB/s which is OK.

I think the desktop machine with a quad core cpu and 8GB Ram is good enough for Samba to perform file transfers at the 60-80MB/s speeds that I am used to seeing, but this little kirkwood box just will not work as fast.

I am in the process of setting up and using NFS instead, but since I have never used it before it is not working well yet.
The shared folder for the 6TB-A is visible on my CentOS when doing the mount, but it shows empty.

Still looking aorund the web to find a fix. Most likely something with folder permissions.

Any suggestions would help - LOL!

"The shared folder for the 6TB-A is visible on my CentOS when doing the mount, but it shows empty."

Most likely it's permission.

-bodhi
===========================
Forum Wiki
bodhi's corner (buy bodhi a beer)

As was already talked about in another thread just a little while ago, most of the network transfer stuff with a kirkwood is going to be quite slow these days because of the encryption algorithms in place for most file transfer protocols, samba and NFS included. None of them are supported by the kirkwood's hardware encryption acceleration onboard so far as I understand, and even at that, nothing makes use of that to my knowledge these days either. Because of this, the kirkwood has to handle encryption in real time with CPU cycles, and for an SoC with no hardware FPU, it cuts into transfer speeds as resources must be used to encrypt/decrypt packets for transfer.

NFS v4 does authentication and encryption through kerberos v5. You might manage maybe 30-35MB/s from what I understand, but that's with a fast kirkwood. speed goes up as SoC clock speed goes up.

But aside, if this drive is using NTFS, you shouldn't. You should be using ExFAT instead for interoperability, in my opinion. ntfs-3g runs in userspace, not kernelspace and is quite slow otherwise, so absolutely anything else will be faster in the long run. I'd also recommend you double-check the drive you have isn't using SMR, which wouldn't be helping anything either.

@sudos,

> because of the encryption algorithms in
> place for most file transfer protocols, samba and
> NFS included. None of them are supported by the
> kirkwood's hardware encryption acceleration
> onboard so far as I understand, and even at that,
> nothing makes use of that to my knowledge these
> days either.

As far as the Kirkwood kernel is concerned, marvell_cesa engine is running (this has been in mainlin since long ago):

[   26.668692] marvell-cesa f1030000.crypto: CESA device successfully registered

Perhaps somebody should rerun the benchmarks like we had before:

Quote
https://forum.doozan.com/read.php?2,23630
Hardware Cryptography

Marvell CESA (also see correction post in this thread)
Marvell CESA in kernel 4.4 performance
Hardware Cryptography cryptodev/openssl On arm5/Debian (build circa 2019)

And then find out which encryption algorigthm, if any, NFS and/or SAMBA is using.

> But aside, if this drive is using NTFS, you
> shouldn't. You should be using ExFAT instead for
> interoperability, in my opinion.

+1

-bodhi
===========================
Forum Wiki
bodhi's corner (buy bodhi a beer)

@Bodhi

I decided to start over and make several changes.

The Goflex will not recognize any of the newer USB sticks that I have, so I was using a very old 4GB Kingston DataTraveler. That thumb drive has seen over 15 years of use so it seemed a bit slow.

Installed a fresh rootfs on the first partition of the 6TB drive, making it a total of 6GB

Formatted the second partition using XFS instead of NTFS - installed xfsprogs

The hdparm speed test for the drive shows slightly faster now that it is using XFS instead of NTFS.

root@debian:/etc# hdparm -tT /dev/sda

/dev/sda:
 Timing cached reads:   514 MB in  2.00 seconds = 256.88 MB/sec
 Timing buffered disk reads: 320 MB in  3.01 seconds = 106.29 MB/sec

Did not install Samba or ntfs-3g, as they are no longer needed.

It seems that both 'nfs-kernel-server' and 'nfs-common' were already installed, so only needed to get 'autofs'

Followed same steps to create swap file and shrink initramfs

installed fake-hwclock before running the apt update and upgrade

Installed hdparm and set 'spindown_time = 60' in the /etc/hdparm.conf file

Made the /etc/sysctl.d/local.conf file with the commands to increase the free page cache, set swapiness to 10, and disable IPV6

Installed cpufrequtils

With only the hard drive connected, sda1 is the roofs and the XFS data partition is sda2

In fstab sda2 is mounted to /mnt/drives/6TB-A using 'noatime'

The /etc/exports file shows the following:

/mnt/drives/6TB-A *(rw,async,insecure,no_root_squash,no_subtree_check,fsid=0,nohide)

On the CentOS machine I make sure the Goflex is sharing the correct folder:

[root@server ~]# showmount --exports 192.168.1.8
Export list for 192.168.1.8:
/mnt/drives/6TB-A *

Then mounted the drive to the home folder (home/server/net)

When using the GUI to copy files, it shows only 15MB/s

If running the copy command from CentOS terminal window, I see the following:

[root@server 4TB]# time cp bigfile /home/server/net

real	0m35.140s
user	0m0.009s
sys	0m0.902s

I am at a loss to understand why this is so slow.

Any ideas would be welcome.

After giving some thought to the issue, there are a few thigs to go over or ask.

As per the Toshiba website, only their consumer and surveillance drives have SMR. This is an enterprise drive and there is no indication that I could find that this drive has SMR.

However, just to be sure I ran hdparm and according to a web posting I found, all drives that use SMR have a feature called 'TRIM'

root@debian:~# hdparm -I /dev/sda | grep TRIM
root@debian:~#

So if it comes back blank, then is most likely does not have SMR.

The decision to use XFS was based on several writeups in web sites indicating that this is one of the most stable and best performance file systems available, especially for files that are more than several hundred megabytes. Since this would be hosting 4K movies in the high gigabytes, then I considered this the best to use for this drive.

If ntfs-3g operates in the userspace and not the kernel space, causing file transfers to be slower, does this also apply to xfsprogs for XFS?

What about ExFat?

In the tests that I have run, it seems the ethernet port and the hard drive itself are performing much higher rates than what is seen when actually transferring a file. Even with the buffered disk reads the disk is showing over 100MB/s.
If the kernel has the Marvell crypto as mentioned by Bodhi (I have the same line when loading the kernel) then why would transfers of actual files be several factors slower than the actual functioning speeds of the hardware?

After becoming interested in using one of these small pogo boxes to host my files, I decided that the GoFlex Net with the two SATA ports was the way to go, considering it also has a cpu running at 1Ghz.
Is there something better that provides higher transfer speeds? (preferably in a similar price range and power usage)

Since I am new to both the forum and also to these devices, I have no experience in what the normal transfer speeds for these kirkwood/pogo boxes should have. Aside from the quite old postings in the Performance Tuning and Benchmarks posts, no one has really mentioned what speeds they are getting with their devices.

Is 15 to 20 MB/s is standard for these devices? What are you guys seeing?

Again, any info or help would be greatly appreciated.

Tedious_1 Wrote:
-------------------------------------------------------
> Since I am new to both the forum and also to these
> devices, I have no experience in what the normal
> transfer speeds for these kirkwood/pogo boxes
> should have. Aside from the quite old postings in
> the Performance Tuning and Benchmarks posts, no
> one has really mentioned what speeds they are
> getting with their devices.
>
> Is 15 to 20 MB/s is standard for these devices?
> What are you guys seeing?
>
> Again, any info or help would be greatly
> appreciated.

If you want to verify it's a speed issue for sure, set up a basic vsftpd install on it with no encryption and try transferring files to it that way and back. Should give you the max throughput you can reasonably expect.

as said, 10-15MB/s is normal when using encrypted filesharing. while cesa exists, programs need to know to specifically work with it. the above test should shed light on where the problem truly lies.

@Tedious_1,

> Is 15 to 20 MB/s is standard for these devices?
> What are you guys seeing?

Have you looked at these 2 threads in the Wiki thread?

Quote

Increase NFSD max_block_size
Reduce NFSD threads

The 1st post about NFS max_block_size. And how to set it.

In the 2nd post davidalfa got about 60MBs read on his Dockstar (the 2nd version), which is basically the same as this Goflex Net in CPU, but RAM is larger, 256MB).

Quote

Tried different max_block_size values.
- Default is 16384. Pretty slow.
- 65536 does a little beter, but not great
- 262144: 54MB/s
- 1048576: 60MB/s.

-bodhi
===========================
Forum Wiki
bodhi's corner (buy bodhi a beer)



Edited 1 time(s). Last edit at 03/26/2024 07:02PM by bodhi.

@ sudos
With all due respect, the idea is to have files transferred to Windows and an Android TV.
I understand the suggestion to see what speeds an unencrypted transfer would be, however that defeats the purpose of using this as a NAS.
There are postings of others using a GoFlex Net/Home with OpenWRT, Archlinux, and Debian getting sustained transfers of around 60MB/s using Samba and NFS.

Quote
bodhi
(THREAD - Linux Kernel 6.7.5 Kirkwood package and Debian rootfs)
Samba and nfs really need to have proper tuning to reach max potentials. I doubt we can say there is a different between Debian and Arch. I think it is actually a difference in tuning, if there is any tuning going on. My released kernel build and rootfs are basic, bare bone, and have no special tuning at all :)

I am interested in seeing how others have 'tuned' their installs

Quote
almaz
(same thread)
I totally agree with Bodhi. My benchmarks were between PogoPlug and Windows PC. That's what is the most important, network speed. I didn't use dd command for testing just evarage what Windows 8.1 graph showed during transfer.

@Bodhi

I followed the suggestions to change the max block size, and did see an increase in speeds, however it quickly ramps down and is not sustained.

Quote
davidfpha
nfsd crashed randomly with socket errors, lowered network window size to 262144 and nfs block size to 524288, seems ok now.

These are the numbers that I stuck to. However when copying a 2Gb mkv video file to and from the CenOS 7 server, the file transfer started off at around 60MB/s but within 10 seconds ramped down to 20.

Also, the suggestion to change the threads does not work.

If you stop the service, write the value, then start the service again, THREADS goes back to 8.

		/etc/init.d/nfs-kernel-server stop
                echo 1 > /proc/fs/nfsd/threads
		/etc/init.d/nfs-kernel-server start


		cat /proc/fs/nfsd/threads
		8

Something seems to be overwriting the variable when starting the service again.

I am hoping someone who has one of these GoFlex Net boxes can pass along their tips and what they use to get the most from this little beast.

Any info, as always, is greatly appreciated.

@Tedious_1,

> Also, the suggestion to change the threads does
> not work.

That was a temporary change.

To make it permanent, change the default settings in /etc/default/nfs-kernel-server

# Number of servers to start up
RPCNFSDCOUNT=1

Restart the service and verify

/etc/init.d/nfs-kernel-server restart
cat /proc/fs/nfsd/threads

-bodhi
===========================
Forum Wiki
bodhi's corner (buy bodhi a beer)



Edited 1 time(s). Last edit at 04/02/2024 06:54PM by bodhi.

Quote

These are the numbers that I stuck to. However when copying a 2Gb mkv video file to and from the CenOS 7 server, the file transfer started off at around 60MB/s but within 10 seconds ramped down to 20.

Try again after you set the NFS thread. And check the swap size while it is running. If there was any increase in swap size, set the swapiness to 1.

-bodhi
===========================
Forum Wiki
bodhi's corner (buy bodhi a beer)



Edited 1 time(s). Last edit at 04/02/2024 06:58PM by bodhi.