Bekijk de volledige versie : USB transfer rate
Hi,
I know about the theoretical value of the USB1.1 transfer rate and want to ask for real life rates of the WL500g, esp. to USB-hardrives?
And what is the overall performance of a WL500g "file server":
With samba?
With ftp (I found one post stating 400-500kB)?
I think it could be of global interest. Thanks in advance.
Peter
JOCKYW2001
08-12-2004, 11:08
I have the same question. Oleg wrote that performance is VERY poor with usb drives. I'd personally be satisfied if it can do about 8-9mbs NFS R/W.
Did anyone actually try?
Hi
My FTP speed to a Flashdrive is around 600 to 700kb, so this is quite good.
The Samba speed was a bit poorer if it worked at all, but this can be a fault of my crapy SMB configuration at the PC.
I can to a little benchmark when im back home.
Greets
If you do a search on the forum you will find the answers. Benchmarks with different firmwares have been posted before.
Samba and FTP speeds (R/W) don't differ much and are around 600 - 700 KByte/s. Do not expect this to change much in any firmware to come.
If these speeds aren't good enough wait for the WL-500gx to arrive or go buy a WL-HDD (There's a new WL-HDD on the horizon, it too should perform better).
/Edit: Removed some nonsense...
...around 600 - 700 KByte/s. Do not expect this to change much in any firmware to come. Oleg proved this speed is CPU limited.
CPU Limited ???
I dont see any heavy load on my device, and its near enough to the USB maximum to be an overhead of the SCSI-Emulation.
Greets
600-700KB/s is the good thing for the USB 1.1. Yes, CPU limits maximum ethernet thruput to something like 3.5MB/s.
Oops, yes, CPU is not limiting the transfer on WL-500g, USB does. Slight difference, same result :) ;)
JOCKYW2001
08-12-2004, 13:34
600-700KB/s is the good thing for the USB 1.1.
Maybe NFS is faster. I'll try it out when I get my hands on an external HDD
Maybe NFS is faster. I'll try it out when I get my hands on an external HDD
That will make no/little difference. Using the FTP server the transferrate is also around 600-700 Kbyte/s.
Perhaps it will fix some other issues with Samba...
Antiloop
08-12-2004, 14:22
If you do a search on the forum you will find the answers. Benchmarks with different firmwares have been posted before.
Samba and FTP speeds (R/W) don't differ much and are around 600 - 700 KByte/s. Do not expect this to change much in any firmware to come.
If these speeds aren't good enough wait for the WL-500gx to arrive or go buy a WL-HDD (There's a new WL-HDD on the horizon, it too should perform better).
/Edit: Removed some nonsense...
a new WL-HDD ? the WL-HDD3.5 I told you about?
or did I miss something :D
a new WL-HDD ? the WL-HDD3.5 I told you about?
or did I miss something :D
Uhuh / Yes / no
Please decode you answer. :D
JOCKYW2001
08-12-2004, 15:14
Improving network performance is haunting me :D
Just found this article:
http://wwwx.cs.unc.edu/~sparkst/howto/network_tuning.php
The full busybox does have sysctl so it's possible to tweak values such as rmem_default or rmem_max
Did anyone try this already?
Well, you do not need sysctl for this. Just an echo to /proc/sys/kernel/ etc...
I've played with this - no significiant differences.
On the other hand I was able to improve performance by something like 10% by reducing number of interrupts on transmission. But this is too low anyway.
Please decode you answer. :D
Okay, I'll start decoding now:
a new WL-HDD ?Uhuh
the WL-HDD3.5 I told you about?Yes
or did I miss something :Dno
Done! ;)
Looks like I've missed something... I've not heard anything about new wl-hdd to be in production. :)
Well, Antiloop told me about it (Don't know if it was confidential or not BTW :rolleyes: ). Anyways, I don't believe its in production allready, its more like an announcement. Antiloop will probably be able to tell you more about this...
Actually its old news:
http://wl500g.info/showpost.php?p=7209&postcount=57
JOCKYW2001
08-12-2004, 16:00
Well, you do not need sysctl for this. Just an echo to /proc/sys/kernel/ etc...
I've played with this - no significiant differences.
On the other hand I was able to improve performance by something like 10% by reducing number of interrupts on transmission. But this is too low anyway.
I tried echo but it didn't work, sysctl worked fine.
(e.g. echo 65536 /proc/sys/net/core/rmem_default)
That 10% gain might be just enough for my purposes. What exactly did you do?
echo 65536 > /proc/sys/net/core/rmem_default
Just modified hnddma to not issue interrupt on then end of each frame, instead changed it to use threshold value for number of tx ring entries available.
The function of interest is dma_txfast (if I recall correctly). But the code was just to try out the thing and it slowed down udp transmission for some reason, seems like I've missed something. So, I've removed this code - 10% is still too slow for me. I still have no idea, why wl500gx works with the same terrible speed. And uses 100% of cpu...
You will need to recompile kernel in order for things to take effect. Recompiling et.o is not enough.
The function of interest is dma_txfast (if I recall correctly). But the code was just to try out the thing and it slowed down udp transmission for some reason, seems like I've missed something. So, I've removed this code - 10% is still too slow for me. I still have no idea, why wl500gx works with the same terrible speed. And uses 100% of cpu...Oleg, you've done tests with WL-500gx allready?
Oleg, you've done tests with WL-500gx allready?
whao performed: http://wl500g.info/showpost.php?p=7633&postcount=21
Damn, how is it possible that a CPU which is more effective and running at a higher speed is only capable of producing the same throughput? :mad:
Would it be a weird idea to suspect something else then the CPU to be the problem? Like onboard busses (CPU <-> Memory or CPU <-> Ethernet)....
Lets hope Antiloop can give us some more info, he should be testing the gx at this moment :cool:
Damn, how is it possible that a CPU which is more effective and running at a higher speed is only capable of producing the same throughput? :mad:
Would it be a weird idea to suspect something else then the CPU to be the problem? Like onboard busses (CPU <-> Memory or CPU <-> Ethernet)....
Lets hope Antiloop can give us some more info, he should be testing the gx at this moment :cool:I don't think this is a weird suggestion, similar suggestions were made by other people before, for example http://wl500g.info/showthread.php?p=6584#post6584 and in also this thread http://wl500g.info/showthread.php?t=1168&highlight=transfer+speed.
I still think that also the WL-500g should be able to reach higher wireless transfer speeds.
I'm very curious what Antiloops WL-500gx findings will be though!!
Ok, so lame Linux kernel drivers can also be a source for this problem, as it seems some pio mode hack is being used to do DMA. Performance can probably be enhanced by optimizing the drivers. Does anyone have any ideas if these drivers are still actively maintained?