[Samba] OS X slow with Samba server

[Jeff McClure]

I have an Apple iBook running OS X 10.2 connected to my network via an 
802.11b wireless network. When I transfer files via SMB with my Windows 
2000 computer, the speed seems reasonable. However, when I connect to a 
share on my Debian (x86) box (kernel 2.4.19) running Samba 2.2.3a-12 for 
Debian, the speed is only about a sixth as fast (transferring files to 
or from the Win2k box is about 6 times faster).

I did some packet captures in the hope of figuring out something. First, 
here's a piece of a transfer from the Win2k box (voga.jeffienet) to the 
iBook (ibook.jeffienet). Sorry about the long lines.

     870 10.080985   voga.jeffienet        ibook.jeffienet       NBSS 
NBSS Continuation Message
     871 10.084876   voga.jeffienet        ibook.jeffienet       NBSS 
   NBSS Continuation Message
     872 10.086621   voga.jeffienet        ibook.jeffienet       NBSS 
   NBSS Continuation Message
     873 10.090163   voga.jeffienet        ibook.jeffienet       NBSS 
   NBSS Continuation Message
     874 10.093957   voga.jeffienet        ibook.jeffienet       NBSS 
   NBSS Continuation Message
     875 10.094042   ibook.jeffienet       voga.jeffienet        TCP 
   49209 > netbios-ssn [ACK] Seq=1308141088 Ack=287400716 Win=65535 Len=0
     876 10.098098   voga.jeffienet        ibook.jeffienet       NBSS 
   NBSS Continuation Message
     877 10.099777   voga.jeffienet        ibook.jeffienet       NBSS 
   NBSS Continuation Message
     878 10.101667   voga.jeffienet        ibook.jeffienet       NBSS 
   NBSS Continuation Message
     879 10.105350   voga.jeffienet        ibook.jeffienet       NBSS 
   NBSS Continuation Message
     880 10.109083   voga.jeffienet        ibook.jeffienet       NBSS 
   NBSS Continuation Message
     881 10.111019   voga.jeffienet        ibook.jeffienet       NBSS 
   NBSS Continuation Message
     882 10.111128   ibook.jeffienet       voga.jeffienet        TCP 
   49209 > netbios-ssn [ACK] Seq=1308141088 Ack=287409404 Win=65535 Len=0
     883 10.115104   voga.jeffienet        ibook.jeffienet       NBSS 
   NBSS Continuation Message
     884 10.117015   voga.jeffienet        ibook.jeffienet       NBSS 
   NBSS Continuation Message
     885 10.120661   voga.jeffienet        ibook.jeffienet       NBSS 
   NBSS Continuation Message
     886 10.122415   voga.jeffienet        ibook.jeffienet       NBSS 
   NBSS Continuation Message
     887 10.126060   voga.jeffienet        ibook.jeffienet       NBSS 
   NBSS Continuation Message
     888 10.126142   ibook.jeffienet       voga.jeffienet        TCP 
   49209 > netbios-ssn [ACK] Seq=1308141088 Ack=287416644 Win=65535 Len=0


Look closely at the packet times. Voga sends a series of data packets, 
then ibook ACKs that group of packets. Notice that the ACK comes almost 
immediately after the last data packet.

Now...take a look at a capture fragment for a file being sent from the 
Linux box (castrovalva.jeffienet) to the iBook:

     152 3.216323    castrovalva.jeffienet ibook.jeffienet       NBSS 
   NBSS Continuation Message
     153 3.219965    castrovalva.jeffienet ibook.jeffienet       NBSS 
   NBSS Continuation Message
     154 3.223540    castrovalva.jeffienet ibook.jeffienet       NBSS 
   NBSS Continuation Message
     155 3.227242    castrovalva.jeffienet ibook.jeffienet       NBSS 
   NBSS Continuation Message
     156 3.230733    castrovalva.jeffienet ibook.jeffienet       NBSS 
   NBSS Continuation Message
     157 3.230805    ibook.jeffienet       castrovalva.jeffienet TCP 
   49208 > netbios-ssn [ACK] Seq=3113175768 Ack=1353220046 Win=65535 Len=0
     158 3.236769    castrovalva.jeffienet ibook.jeffienet       NBSS 
   NBSS Continuation Message
     159 3.238683    castrovalva.jeffienet ibook.jeffienet       NBSS 
   NBSS Continuation Message
     160 3.242363    castrovalva.jeffienet ibook.jeffienet       NBSS 
   NBSS Continuation Message
     161 3.246103    castrovalva.jeffienet ibook.jeffienet       NBSS 
   NBSS Continuation Message
     162 3.248020    castrovalva.jeffienet ibook.jeffienet       NBSS 
   NBSS Continuation Message
     163 3.404769    ibook.jeffienet       castrovalva.jeffienet TCP 
   49208 > netbios-ssn [ACK] Seq=3113175768 Ack=1353227286 Win=59332 Len=0
     164 3.410756    castrovalva.jeffienet ibook.jeffienet       NBSS 
   NBSS Continuation Message
     165 3.410908    ibook.jeffienet       castrovalva.jeffienet TCP 
   49208 > netbios-ssn [ACK] Seq=3113175768 Ack=1353228734 Win=65535 Len=0
     166 3.412829    castrovalva.jeffienet ibook.jeffienet       NBSS 
   NBSS Continuation Message
     167 3.416726    castrovalva.jeffienet ibook.jeffienet       NBSS 
   NBSS Continuation Message
     168 3.420294    castrovalva.jeffienet ibook.jeffienet       NBSS 
   NBSS Continuation Message
     169 3.422117    castrovalva.jeffienet ibook.jeffienet       NBSS 
   NBSS Continuation Message
     170 3.605023    ibook.jeffienet       castrovalva.jeffienet TCP 
   49208 > netbios-ssn [ACK] Seq=3113175768 Ack=1353234526 Win=60284 Len=0
     171 3.610969    castrovalva.jeffienet ibook.jeffienet       NBSS 
   NBSS Continuation Message
     172 3.612829    castrovalva.jeffienet ibook.jeffienet       NBSS 
   NBSS Continuation Message
     173 3.612910    ibook.jeffienet       castrovalva.jeffienet TCP 
   49208 > netbios-ssn [ACK] Seq=3113175768 Ack=1353237422 Win=65535 Len=0

Again, look closely at the relative packet times. This time, it appears 
that ibook waits something like 180ms before it acks the group of data 
packets. This is only a small snippet, but the pattern is consistent 
throughout the transfer. When talking to the Debian Samaba server, the 
iBook takes MUCH MUCH longer to send a lot of the ACKs. That accounts 
for the slower transfer.

Now...the $64,000 question: what makes the iBook behave differently with 
the Samba server than with the Win2k box? This is really frustrating, 
because SMB really is the best way for me to move files around. The 
authentication mechanism in NFS is braindead, and AppleTalk (using 
Netatalk on the Linux box) doesn't allow filenames >31 bytes long. If I 
can solve this speed problem, I'll have it made.

I will be happy to collect more data if someone can provide an 
intelligent suggestion.

--Jeff McClure