2.6.22/realtek bug in hardware, any kernel work-around?

2.6.22/realtek bug in hardware, any kernel work-around?

[Justin Piszcz]

Package: samba
Version: 3.0.26a-1
>
> Kernel: 2.6.22
>
> samba 3.0.26a-1 performance < 900 KiB/s, but FTP = 30-90 MiB/s
>
> Let me start out by saing this is an oddball problem:
>
> SAMBA:
> LINUX -> WINDOWS = < 900 KiB/s (varies between 100 - 900 KiB/s)
> WINDOWS -> LINUX = 30-90 MiB/s (always)
>
> FTP:
> Either direction, 30-90 MiB/s (always)
>
> I do not see any nasty errors in the logs even with verbose = 5.
>
> Any ideas here? I am not using any special options.
>
> # cat /etc/samba/smb.conf
>
> [global]
>        log level = 5
>        workgroup = WORKGROUP
>        server string = %h - Pentium IV 3.4GHZ
>        security = user
>        encrypt passwords = true
>
> [user]
>  comment         = user
>  path            = /home/user
>  writable        = yes
>  valid users     = user
>  create mask     = 644
>
> --
>
> Here, FTP for pulling files from Linux.
>
> ftp> mget *
> 200 TYPE is now 8-bit binary
> 200 PORT command successful
> 150-Connecting to port 1255
> 150 715924.0 kbytes to download
> 226-File successfully transferred
> 226 13.687 seconds (measured here), 51.08 Mbytes per second
> ftp: 733106176 bytes received in 13.69Seconds 53562.23Kbytes/sec.
> 200 PORT command successful
> 150-Connecting to port 1256
> 150 716272.0 kbytes to download
> 226-File successfully transferred
> 226 13.032 seconds (measured here), 53.67 Mbytes per second
> ftp: 733462528 bytes received in 13.05Seconds 56216.95Kbytes/sec.
> 200 PORT command successful
> 150-Connecting to port 1257
> 150 713200.0 kbytes to download
> 226-File successfully transferred
> 226 12.869 seconds (measured here), 54.12 Mbytes per second
> ftp: 730316800 bytes received in 12.88Seconds 56723.63Kbytes/sec.
>
> Here, FTP for pushing files to Linux.
>
> ftp> mput 1 2 3
> 200 PORT command successful
> 150 Connecting to port 1263
> 226-File successfully transferred
> 226 12.802 seconds (measured here), 54.61 Mbytes per second
> ftp: 733106176 bytes sent in 12.80Seconds 57287.35Kbytes/sec.
> 200 PORT command successful
> 150 Connecting to port 1264
> 226-File successfully transferred
> 226 12.949 seconds (measured here), 54.02 Mbytes per second
> ftp: 733462528 bytes sent in 12.95Seconds 56624.92Kbytes/sec.
> 200 PORT command successful
> 150 Connecting to port 1265
> 226-File successfully transferred
> 226 15.400 seconds (measured here), 45.23 Mbytes per second
> ftp: 730316800 bytes sent in 15.38Seconds 47500.28Kbytes/sec.
>
> But (all I can offer is packet dumps/traces or bandwidth measurements):
>
> Incoming:                               Outgoing:
> Curr: 0.00 MByte/s                      Curr: 0.07 MByte/s
> Avg: 0.00 MByte/s                       Avg: 0.07 MByte/s
> Min: 0.00 MByte/s                       Min: 0.07 MByte/s
> Max: 0.00 MByte/s                       Max: 0.07 MByte/s
> Ttl: 1898.08 MByte                      Ttl: 2954.92 MByte
>
> LOCAL <-> REMOTE                                          TXBPS   RXBPS 
> TOTALBPS
> (IP)              PORT  PROTO  (IP)              PORT       TX      RX 
> TOTAL
> linuxbox <-> p4w.internal.lan          546k/s 4.74k/s 551k/s
> 192.168.0.1     445    TCP  192.168.0.2    1259    6.88m    106k  6.99m
>
> Why do I get such poor performance when trying to retrieve a file off the 
> Linux box?  This is a very strange problem.
>
> Linux:
>
> $ netstat -i
> Kernel Interface table
> Iface   MTU Met   RX-OK RX-ERR RX-DRP RX-OVR    TX-OK TX-ERR TX-DRP TX-OVR 
> Flg
> eth0   1500 0  14049682      0      0      0 11354070      0      0      0 
> BMRU
> lo    16436 0     45335      0      0      0    45335      0      0      0 
> LRU
>
> Windows:
>
> netstat -s
>
> IPv4 Statistics
>
>  Packets Received                   = 5053597
>  Received Header Errors             = 0
>  Received Address Errors            = 19
>  Datagrams Forwarded                = 0
>  Unknown Protocols Received         = 0
>  Received Packets Discarded         = 2
>  Received Packets Delivered         = 5053595
>  Output Requests                    = 3655144
>  Routing Discards                   = 0
>  Discarded Output Packets           = 0
>  Output Packet No Route             = 0
>  Reassembly Required                = 0
>  Reassembly Successful              = 0
>  Reassembly Failures                = 0
>  Datagrams Successfully Fragmented  = 0
>  Datagrams Failing Fragmentation    = 3
>  Fragments Created                  = 0
>
> ICMPv4 Statistics
>
>                            Received    Sent
>  Messages                  47          24
>  Errors                    0           0
>  Destination Unreachable   25          2
>  Time Exceeded             0           0
>  Parameter Problems        0           0
>  Source Quenches           0           0
>  Redirects                 0           0
>  Echos                     0           22
>  Echo Replies              22          0
>  Timestamps                0           0
>  Timestamp Replies         0           0
>  Address Masks             0           0
>  Address Mask Replies      0           0
>
> TCP Statistics for IPv4
>
>  Active Opens                        = 204
>  Passive Opens                       = 14
>  Failed Connection Attempts          = 16
>  Reset Connections                   = 59
>  Current Connections                 = 3
>  Segments Received                   = 5051709
>  Segments Sent                       = 3654327
>  Segments Retransmitted              = 76
>
> UDP Statistics for IPv4
>
>  Datagrams Received    = 1826
>  No Ports              = 58
>  Receive Errors        = 3
>  Datagrams Sent        = 716
>
>
> Justin.
>
> -- 
> To unsubscribe from this list go to the following URL and read the
> instructions:  https://lists.samba.org/mailman/listinfo/samba
>

Looks like it is hardware related after all:

http://forums.gentoo.org/viewtopic-p-2820556.html
http://ubuntuforums.org/showthread.php?t=531505
http://www.fedoraforum.org/forum/showthread.php?p=452451

One poster wrote "I got similar problems:"

http://forums.gentoo.org/viewtopic-t-570392.html
http://forums.gentoo.org/viewtopic-t-389449-postdays-0-postorder-asc-start-25.html?sid=4251f92bbc6db32fd200c4ea0ea9c9be

I found out that poor smb performance is related to a broken udp-perfomance of the 8169 NIC in outgoing udp.
But, only in Gb Mode of the 8169 NIC, with 100Mbps it is ok.

http://forums.gentoo.org/viewtopic-t-570392.html

I have:
03:00.0 Ethernet controller: Realtek Semiconductor Co., Ltd. RTL8111/8168B PCI Express Gigabit Ethernet controller (rev 01)
04:00.0 Ethernet controller: Realtek Semiconductor Co., Ltd. RTL8111/8168B PCI Express Gigabit Ethernet controller (rev 01)

That must be the issue, time to buy an Intel PCI-e NIC if I want better 
performance with samba in Linux.

I do have another one of these motherboards (ABIT AW9D-MAX) running XP and 
it does not seem to suffer from this problem, I guess the [proprietary? 
Realtek driver does not suffer from this bug]?

Justin.

Re: [Samba] 2.6.22/realtek bug in hardware, any kernel work-around?

[Sinisa Bandin]

Justin Piszcz wrote:
> Package: samba
> Version: 3.0.26a-1
>>
>> Kernel: 2.6.22
>>
>> samba 3.0.26a-1 performance < 900 KiB/s, but FTP = 30-90 MiB/s
>>
>> Let me start out by saing this is an oddball problem:
>>
>> SAMBA:
>> LINUX -> WINDOWS = < 900 KiB/s (varies between 100 - 900 KiB/s)
>> WINDOWS -> LINUX = 30-90 MiB/s (always)
>>
>> FTP:
>> Either direction, 30-90 MiB/s (always)
>>
>> I do not see any nasty errors in the logs even with verbose = 5.
>>
>> Any ideas here? I am not using any special options.
>
> I found out that poor smb performance is related to a broken 
> udp-perfomance of the 8169 NIC in outgoing udp.
> But, only in Gb Mode of the 8169 NIC, with 100Mbps it is ok.
>
I had exactly the same problem with 100Mb/s also.

Solved by disabling onboard LAN interface and inserting (US Robotics) 
Realtek 8139 based card (10/100 Mbps).

Best regards,
Sinisa Bandin

Re: [Samba] 2.6.22/realtek bug in hardware, any kernel work-around?

[Justin Piszcz]

On Sun, 30 Sep 2007, Sinisa Bandin wrote:

>
> Justin Piszcz wrote:
>> Package: samba
>> Version: 3.0.26a-1
>>> 
>>> Kernel: 2.6.22
>>> 
>>> samba 3.0.26a-1 performance < 900 KiB/s, but FTP = 30-90 MiB/s
>>> 
>>> Let me start out by saing this is an oddball problem:
>>> 
>>> SAMBA:
>>> LINUX -> WINDOWS = < 900 KiB/s (varies between 100 - 900 KiB/s)
>>> WINDOWS -> LINUX = 30-90 MiB/s (always)
>>> 
>>> FTP:
>>> Either direction, 30-90 MiB/s (always)
>>> 
>>> I do not see any nasty errors in the logs even with verbose = 5.
>>> 
>>> Any ideas here? I am not using any special options.
>> 
>> I found out that poor smb performance is related to a broken udp-perfomance 
>> of the 8169 NIC in outgoing udp.
>> But, only in Gb Mode of the 8169 NIC, with 100Mbps it is ok.
>> 
> I had exactly the same problem with 100Mb/s also.
>
> Solved by disabling onboard LAN interface and inserting (US Robotics) Realtek 
> 8139 based card (10/100 Mbps).
>
> Best regards,
> Sinisa Bandin
>

Thank you for your response, so the bottom line is Realtek 1Gbps NICs are 
unreliable.  Wonder what Realtek has to say about it if they even respond, 
added them to cc list.

Justin.

Re: [Samba] 2.6.22/realtek bug in hardware, any kernel work-around?

[Sinisa Bandin]

Justin Piszcz wrote:
>
>
> On Sun, 30 Sep 2007, Sinisa Bandin wrote:
>
>>
>> Justin Piszcz wrote:
>>> Package: samba
>>> Version: 3.0.26a-1
>>>>
>>>> Kernel: 2.6.22
>>>>
>>>> samba 3.0.26a-1 performance < 900 KiB/s, but FTP = 30-90 MiB/s
>>>>
>>>> Let me start out by saing this is an oddball problem:
>>>>
>>>> SAMBA:
>>>> LINUX -> WINDOWS = < 900 KiB/s (varies between 100 - 900 KiB/s)
>>>> WINDOWS -> LINUX = 30-90 MiB/s (always)
>>>>
>>>> FTP:
>>>> Either direction, 30-90 MiB/s (always)
>>>>
>>>> I do not see any nasty errors in the logs even with verbose = 5.
>>>>
>>>> Any ideas here? I am not using any special options.
>>>
>>> I found out that poor smb performance is related to a broken 
>>> udp-perfomance of the 8169 NIC in outgoing udp.
>>> But, only in Gb Mode of the 8169 NIC, with 100Mbps it is ok.
>>>
>> I had exactly the same problem with 100Mb/s also.
>>
>> Solved by disabling onboard LAN interface and inserting (US Robotics) 
>> Realtek 8139 based card (10/100 Mbps).
>>
>> Best regards,
>> Sinisa Bandin
>>
>
> Thank you for your response, so the bottom line is Realtek 1Gbps NICs 
> are unreliable.  Wonder what Realtek has to say about it if they even 
> respond, added them to cc list.
>
> Justin.
>
Wouldnt go so far to say that they are unreliable, but there is certanly 
some problem between Samba and 8169 driver (and maybe other driver, 
don't remember for sure, but will check first thing in the morning)

All other protocols work perfectly, I have been able to use full 1 Gbps 
with only two concurent transfers with all of scp, rsync, http and ping 
-f -s 32000 (and any conbination of those).


Sinisa

Re: [Samba] 2.6.22/realtek bug in hardware, any kernel work-around?

[Francois Romieu]

Justin Piszcz <jpiszcz@lucidpixels.com> :
[...]

The bug is fixed in 2.6.23-rc9. Try it.

-- 
Ueimor