SMB Server and slow transfer speeds...
 

I am having SERIOUS file transfer issues from a Linux SMB server to a Mac running OS-X.

My setup:

Win2k Domain / DB / Exchange Server
Linux File / Web / Backup Server
30 Win2k Clients
1 OS-X Client
2 Netgear DS524 Hubs

I get great performance on all accounts from the file server with the windows clients.

When the Mac user tries to transfer files from the server, it is painfully slow. I've commented out the references to the send and recieve buffer size, but that makes no difference.

I would like to get to the bottom of this, because it also appears to be affecting it's Outlook / Exchange traffic. It is not unusual for the Classic environment to "lag" periodically during the day. The report I get is that it is somewhat smooth in the AM...

This sucks.

JC


__________________
------------------------------------------
Certified LinuxZealot
11:25:54 up 14965 days, 22:26, 1 user, load average: 1.27, 1.29, 1.23

ok, the usual networking questions:
How many active networking ports are open on the mac?
check by looking at the network pref pane and looking at the 'Show:' pull down menu. The bottom choice "Network Port Configuration" lets you delete and deactivate extra ports. This turns out to be importantbecause otherwise the mac checks all active ports for sending/recieving each and every packet (or so i am told). and worse comes to worse its a simple first step.
next, chek the cables and the ports, easiest thing to do, if the mac sits near a PC that works fine, just switch the tailing end (not the wall side, but the comp side) of the network cables, that in effect changes both cbale and port in one easy shot. then test both machiens. if the pc slows down and/or the mac speeds up, you have found a major portion of your problem.
Next, how much ram does the machine have? How many extra little programs ahve been put on it that use up bandwith/CPU/RAM? I had a client who spent three hours on the phone with me complaining of slow email and file transfers and neglected to tell me about the 30 or so extra doodads running on the machine. When i explained that each one, the weather checker, the three chat clients, the P2P client, the screensaver as desktop pattern, the little barking doggy, the US flag, the dancing girl, etc. ... that each one takes up just a bit of ram and cpu and bandwith, and that all together they kill the machine, the user said "OH, then i guess when i installed them all yesterday, that would explain why things slowed down" yes, however the user was an artist and refused to get rid of them so I had no good way of speeding up the machine.
ie, check for exta crap and enough ram.

also, just cause it can't hurt, when was the last time you ran permission fixer (in disk utility)? And out of curiosity, have the daily/weekly/monthly cron jobs been running? ie has the machine been on and awake all night? the cron jobs will only run if the machine is both on and awake at night at like 3 or 4 am. If you don't wnat to leave it on and awake all the time, get ResurrXtion (search versiontracker for it) to turn your machien on or wake it up at the right times.

i think thats a good start.
Knowmad

Well...
 

There is only one port configuration, ethernet. The machine is a dual 877 with 768 mb of ram. Speed is not an ussue. The only apps running that take up excessive bandwidth or CPU usage is Classic and Outlook connecting to the Exchange server. This is clearly an issue with the Samba server and OS-X because I can download files faster from the internet than I can on the local lan. I also changed the cat5 with no diff.

Thanks for all your input.

JC

bad news
 

I looked around and your not the only one, not by a long shot, who is having this dificulty.
I did not, however see any answers.
My suggestions are:
look on google groops
look on AFP548.com
post your findings back here.

sorry i can't help more, i have not yet myself had this problem and without being there I have little I can do.... myabe if you post the log files from the SMB service we might find something?

knowmad

A test...
 

The mac:

[STATION10:/Volumes/company/Technology] marketing# ls -la
total 632907
drwxr-xr-x 1 jzalewsk wheel 16384 Sep 12 08:38 .
drwxr-xr-x 1 jzalewsk wheel 16384 Sep 12 07:42 ..
drwxr-xr-x 1 jzalewsk wheel 16384 Apr 4 14:39 Access articles
drwxr-xr-x 1 jzalewsk wheel 16384 Apr 4 14:39 Alpha ECU
-rwxr-xr-x 1 jzalewsk wheel 315922432 Sep 10 10:23 Dec2002DevToolsCD.dmg.bin
-rwxr-xr-x 1 jzalewsk wheel 782148 Jun 2 08:58 ffbgalmac.zip
-rwxr-xr-x 1 jzalewsk wheel 7279051 Jun 26 11:18 lccinstalldisk111.dmg

[STATION10:/Volumes/company/Technology] marketing# time cp ffbgalmac.zip /Users/administrator/Desktop
0.000u 0.010s 0:04.03 0.2% 0+0k 0+2io 0pf+0w


The Fileserver on loglevel 2:

[2003/09/12 11:25:39, 2] smbd/open.c:open_file(245)
JZALEWSKI opened file Technology/ffbgalmac.zip read=Yes write=Yes (numopen=1)
[2003/09/12 11:25:43, 2] smbd/close.c:close_normal_file(213)
ramart+jzalewski closed file Technology/ffbgalmac.zip (numopen=0)

That is just too long...

I'm going to recommend that she keep her machine on and see what happens.

Just for the record, I had a similar setup at another location, but with an NT fileserver, and never had these issues.

JC

I had this happen once and it got better.
 

I had this happen once and it got better. I reset/rebooted both systems... Not that is your problem ... ** What version of OSX??

----
As a test why not make a new Admin level account on the Mac and try the share thing again and see what happens.

You have more or less eliminated the network and PC so somthing must be up on the mac.

Beyond the new account, Permissions or Dir damage may be an issue too.

Well...
 

The OS is 10.2.6, and rebooting the Mac is no problem. We do that every day. I'm sure the issue is with the Mac, since most articles I've read have indicated that the downstream is affected, not the upstream, which is the case here. I can upload a sizable file to the file server in resonable time, but to download is a pain.

I'll try the new account.

Thanks.

JC

duplexing?
 

Ok, i read somewhere here (past forum post on similar subject, can't find it or i would link to it) that sometimes the macs will miss-auto-negotiate duplex in one direction only, a way around this to set the duplexing and not let it autonegotiate. So, set duplexing and speed, don't let it autonegotiate, you do this the easy way with maintain1 or the hard way by finding and editing the appropriate settinsg files. I suggest getting maintain1 from versiontracker and using many of its scripts often as needed. makes life easy.
I am still waiting for a friend of mine to give me some more insight into your issues, he deasl with mac/linux transfers all the time.
knowmad

Some OS and Some hardware have issues with Auto
 

Some OS and Some hardware have issues with Auto.

I doubt that is the case here, though anything is possible.

New news...
 

My problem appears to be OSX itself. Changed the duplexing option in a shell, no diff. Rebooted into OS9, downloaded and installed an evaluation version of Dave for OS9, and the Mac transferred a large file in the same amount of time the Win2k clients do.

Since that part is isolated, I wonder where to go from here? Is this an OSX bug in the new update? I'm running 10.2.6...

JC

What happended to creating a new account??
 

What happended to creating a new account??

At least you eliminated the NIC on the MAC

I guess...
 

No change. I had an admin account set up long ago, and even checked the *GULP* root account..."bad admin, bad". No diff.

This is pretty bizzar, no?

Re: duplexing?
 

Maybe you were thinking of the following thread?
http://forums.macosxhints.com/showth...=&postid=56740

I also suggest that you look at the output of 'netstat' while doing the slow transfers to see if there are a large number of errors or retransmissions, etc.

I did notice...
 

That I was getting my fair share of bad packets. Maybe 10% or so.

JC

locking is Only Good if both Side Agree
 

Anyway..

Here is how to set Speeds to various Locks including Full Duplex Lock:

http://docs.info.apple.com/article.html?artnum=107475


Good Luck

P.S. Unix is fun but this should be the GUI....

Thanks, but
 

I tried this already in a term. Made things worse, and to top it off, it had autoselected half-duplex in the first place, which was correct. My fileserver nic drivers HATE misconfigured nics on the network (thank you Donald Becker), and log excessively if ANYTHING is set to full-duplex.

I'm glad I've got it isolated to OSX, just have to keep plugging away to find a solution.

JC

diagnostic tool
 

You might try running the diagnostic tool from this web site:
http://miranda.ctd.anl.gov:7123/

It is intended for diagnosing Internet connectivity problems but if you can make your Internet connection go via the same routers that connect you to the file servers, it might prove useful.

Re: Thanks, but
 

Right.
The rule is very simple:
Never Force Full Duplex.
It doesn't solve any problems and usually makes things worse.

Breen

Thanks for the tip.
 

Here are the results:

TCP/Web100 Network Diagnostic Tool v5.2.0f
click START to begin
Checking for Middleboxes . . . . . . . . . . . . . . . . . . Done
running 10s outbound test (client to server) . . . . . 654.77Kb/s
running 10s inbound test (server to client) . . . . . . 456.62kb/s
Your Workstation is connected to a Cable/DSL modem

Looks pretty good to me. I'm not sure were to go from here.

JC

as an FYI Sun Ultra 1 and Ultra 2 running
 

As an FYI Sun Ultra 1 and Ultra 2 running Solaris 2.51 and 2.6 do strongly prefer being locked and @ 100 Full. Not relevant here though:)

---------------

Okay ball of confussion.

Do we have another X mac that is working, as we have already extablshed that all is hunky Dory from this mac in 9.x

Unfortunately...
 

We only have one Mac.

Given my situation, I'll take the optimist tack and call this fortunate...:)

FYI...
 

I just found a great GUI to loads of 'nix shell utils:

http://www.apple.com/downloads/macos...s/xupport.html

It seemed to improve things a tad...

JC

Neat Util
 

Neat Util... As far as the "one mac comment" it should not happen, but strange things happen with all OS, it is what many of us are paid to deal with indirectly or directly. Part of trouble shooting requires testing varibles until one finds the solution.

Do you have access to a spare RAW ata mechanism or a FW drive case. You could do a clean install... If you DO NOT have an extra drive you could do an Archive and Install: Preserving Data and.....

Archive and Install FYI:

There are two Archive and Install Options, with and with out saving Apps and Users Prefs.

They both Take Your OSX Folder the ("System") Folder and Make it into a Previous one. If you choose to Save your Settings It will Save you from user setup where you assign passwords, It will save you from having to Drag things from your Applcaitons Folder to your new one, It will preserve your Desktop, Your Dock etc.

You will have to recreate printers and repatch the OS back to 10.26 etc. and counting, reinstall any custom drivers e.g. Carbon Copy Cloner, If you have any special non apple mice etc. You will have to reinstall Palm to make HotSync work correctly, but your data will still be there.

It is super fast compared to NOT saving your user settings. The ? is will this work with this nebulous problem.

**I would suggest **Not saving user Setting as it will probably propagate the problem.

As always it would be irresponsible if did not tell you to back up your data "just in case" somthing unlikely happend or human error like you do a Format install etc.

------------------------
Good Luck ******************************

I just did that....
 

Before all this happened. Our user was complaining of a number of issues, one of which was network related, so I took an afternoon she was off and started ALL over again.

Re-initialized the HDD. Partitioned it into 3 parts:
OS - 5gb
Apps Swap - 2gb
Apps - remaining 27gb or so

No difference...

I'm copying parts of this thread over to the Apple discussion area. Maybe one of the folks over there know something we don't.

Thanks for the input.

JC

You do not need to Parittion like that anymore
 

You do not need to Partition like that anymore..... for OSX 10.2x and later

Perhaps that is part of thw problem free space on the Boot Partition ... swap file thrashing etc? I would strongly suggest just having one partition.

I am sorry we have not yet unscrambled the egg. I would suggest well keep trying with you.

I realize...
 

that it's not necessary to partition the drive up that way, but I thought it would make things easier in the event I needed to reinstall the OS again. The APPS partition only contains our marketing-type software, Photoshop, Quark, etc. The system swap partition is still located on the primary OS partition, and the 2gb APP partition is only for Photoshop and Illustrators usage.

JC

Thats what Arhcive and Install is for:)
 

Thats what Arhcive and Install is for:)

Still in...
 

my Linux mode...I guess I think too much...

JC

an Appology/suggestion
 

It seems i accidentally hijacked the conversation by mentioning duplexing. I am so sorry. I am going to ask, as part of my appology, that we see if we can help the guy fix the problem.
Knowmad
(thats all i wrote before editting)

Now that i have seen the second page (talk about missing the obvius) I have a suggestion.
You can check to see if it is a lack of hard drive space creating swapfile-death that is casuing your problem without doing a full rebuild, though it will be close.
try the following:
install fresh onto the 27gb partition.
test the transfer speeds before upfating
update the combo
test transfer
update security patches a few at a atime, testing after every group
stop when throughput dies.

IF, as i suspect, it is the partition size you will never see the performance drop. If it is one of the updates, you will know which one, if it is osX you will know before updating and can save yourself a few hours.
tell us what happens.
ps am still waiting for my networking guru friend to get of his rear and answer me re this whole thing.
knowmad

Not at all...
 

It's pretty much stalled at this point...don't sweat it. For what it's worth, you've all been very helpful.

JC

Well...I see your point.
 

These are the drive spaces...

[STATION10:~] marketing% df -kl
Filesystem 1K-blocks Used Avail Capacity Mounted on
/dev/disk0s9 20181700 4010288 15969596 20% /
fdesc 1 1 0 100% /dev
/dev/disk0s10 1048440 132648 915792 12% /Volumes/Scratch Disk
/dev/disk0s11 38815692 1634216 37181476 4% /Volumes/Applications

As you can see, the amount of space used on the primary disk is very little. There appears to be nearly 16gig left.

Forgive my poor memory of drive space allocated before....

JC

crap, i thought i was onto something..... hrm, well the install on other partition by stages is not a terrqable idea, juts time consuming. I will keep thinking.
knowmad

I've considered it...
 

And I may still do it. I have the rest of this week to do what I like to the box before our marketing person gets back from vacation.

I think the idea you have about reinstalling on another partition is very good, since I believe that the problem is related to an update the machine recieved.

I think I read somewhere that the 10.2.6 updated some wireless network files...go figure.

JC

can you post the smb.conf file from the file server?
socket options = IPTOS_LOWDELAY TCP_NODELAY has fixed issues for me before. The line will look like this by default:
socket options = IPTOS_LOWDELAY TCP_NODELAY SO_SNDBUF=4096 SO_RCVBUF=4096

which version of samba is on the linux box? is that box set to auto-neg on the nic card?

have you tried to run a constant ping on the file server? you'd want to check for packet loss. I'd let it run for a good 30 minutes and see what the packet loss is.

10.2.6 has a fix for samba that affected files being corrupted when being copied. might be a couple of other patches in that src tree.

Thanks for the info...
 

I didn't know about the samba patch in the latest 10.2.6 update. Is there a way to "roll back"?

I have tried commenting out those references to send and recieve buffers in my smb.conf to no avail. My original conf increased the i/o buffer twice the default of 4096, and neither of the three changes I made positively affected performance.

I am using David Beckers build of the 3com Hurrican driver. It does autonegoiate correctly to half-duplex, and in my earlier posts, writes copious amounts of complaining to syslog when ANYONE on the local lan is misconfigured. Really keeps me on my toes.

Thanks for the input.

BTW, my build is 2.2.7. Solid as a rock...

JC

can you post your SMB.conf or the line in question at least?

What's the ping look like?

Thanks

Sure...
 

# Global parameters
[global]
workgroup = RAMART
server string = Ramart File Server
security = DOMAIN
encrypt passwords = Yes
obey pam restrictions = Yes
password server = RAMART1
pam password change = Yes
passwd program = /usr/bin/passwd %u
passwd chat = *New*password* %n\n *Retype*new*password* %n\n *passwd:*all*authentication*tokens*updated*successfully*
log level = 1
log file = /var/log/samba/%m.log
max log size = 0

#################################################
# This is the original line. It has been COMPLETELY commented out, and changed to # 4096...

socket options = TCP_NODELAY IPTOS_THROUGHPUT SO_RCVBUF=8192 SO_SNDBUF=8192

#################################################

#This was set to minimize Mrxsmb 3034 Event ID errors on the Win2k server logs
# Check EventID.net for details
os level = 1
domain master = no
local master = no
preferred master = no
#end of add

dns proxy = No
winbind uid = 10000-20000
winbind gid = 10000-20000
template homedir = /home/%U
template shell = /bin/bash
winbind separator = +
admin users = noneofyourbeezwax
csc policy = documents

# Veto MAC .DS_Store file creation, and infectious email and news articles
# Cron job will also prune these files
veto files = /.DS_Store/*.eml/*.nws/*.{*}/.xvpics/desktop.ini/DESKTOP.INI/
hide dot files = true
delete veto files = true

[homes]
comment = %U's Home Directory
read only = No
create mask = 0664
directory mask = 0775
browseable = No

[Company]
path = /home/company
read only = No
create mask = 0664
directory mask = 0775

[Apps]
path = /home/apps
write list = noneofyourbeezwax

[Office 2000]
path = /home/office
write list = noneofyourbeezwax

[Win2k Options]
path = /mnt/options
write list = noneofyourbeezwax

[desktop]
browseable = no
writeable = no
path = /home/desktop
write list = noneofyourbeezwax

[home]
path = /home
write list = noneofyourbeezwax
browseable = no
writeable = no

[modifications]
path = /home/modifications
write list = noneofyourbeezwax
browseable = no
writeable = no

[quickbooks]
path = /home/quickbooks
write list = noneofyourbeezwax
browseable = no
writeable = no

[marketing]
path = /home/marketing
read only = no
create mask = 0664
directory mask = 0775

JC

I hate to be a pain but can you try this?

socket options = TCP_NODELAY IPTOS_THROUGHPUT SO_RCVBUF=8192 SO_SNDBUF=8192

Try just socket options = TCP_NODELAY IPTOS_THROUGHPUT

Then restart the smbd and nmbd(if you're using it) daemons from the init script or by hand.

Try your tests and let me know if it performs the same way.

Re-rolling back from 10.2.6 - I believe the patch is "a good thing" and that you would necess want to roll it back. I'll dig up the details.

Re: I did notice...
 

Very interesting thread you are having.

I think you should look into explaining this, or at least under which circumstances it occurs
10% bad packets is quite a lot in my book. Does anyone know how Mac OS X reacts to this ?

Sorry one more then - I'd also like to see you try :
IPTOS_LOWDELAY TCP_NODELAY as the options. Right now you have TCP_NODELAY IPTOS_THROUGHPUT

In addtion when you comment the lines out make sure you put the hash before the line itself and not before or after it. Better yet remove the lines totally and keep a couple copies of the config fil.

The constant ping output is also going to be very interesting.

No noticeable difference...
 

root@mybox:~# testparm | grep 'socket options'
socket options = TCP_NODELAY IPTOS_NODELAY

5-6 minutes for a 18.1mb download.


[STATION10:~/Desktop] administrator% ping mybox > test.txt

[STATION10:~/Desktop] administrator% cat test.txt | more
PING mybox (no.not.today.dude): 56 data bytes
64 bytes from no.not.today.dude: icmp_seq=0 ttl=64 time=0.325 ms
64 bytes from no.not.today.dude: icmp_seq=1 ttl=64 time=0.277 ms
64 bytes from no.not.today.dude: icmp_seq=2 ttl=64 time=0.285 ms

----------------- SNIP FOR SANITY??? -------------------------------------

64 bytes from no.not.today.dude: icmp_seq=168 ttl=64 time=0.249 ms
64 bytes from no.not.today.dude: icmp_seq=169 ttl=64 time=0.244 ms
64 bytes from no.not.today.dude: icmp_seq=170 ttl=64 time=0.248 ms
--- mybox ping statistics ---
375 packets transmitted, 375 packets received, 0% packet loss
round-trip min/avg/max = 0.198/0.269/0.427 ms


Perfect...

JC

> socket options = TCP_NODELAY IPTOS_NODELAY

No - TCP_NODELAY IPTOS_LOWDELAY

LOWDELAY not NODELAY

Then let's try it.

An additional idea, in my case raising the log level from 0 to 3 will cut the untuned get speed from 645.3 to 622.2KB/s, or roughly 5 percent.

so i have my log level at 1

here's my usual tweaks to smb.conf.

[global]
log level = 1 # Default is 0
socket options = TCP_NODELAY IPTOS_LOWDELAY
read raw = yes # Default
write raw = yes # Default
oplocks = yes # Default
max xmit = 65535 # Default
dead time = 15 # Default is 0
getwd cache = yes
lpq cache = 30
[localseg]
veto oplock files = /var/lock/smb.lock
[remoteseg]
oplocks = no

OK.
 

root@mybox:~# testparm | grep 'socket options'
socket options = IPTOS_LOWDELAY TCP_NODELAY

root@mybox:~# /etc/rc.d/init.d/smb reload
Reloading smb.conf file: [ OK ]

That reduced the transfer time of the 18.1mb file to 2 minutes!

JC

Just to be noteworthy, not picky, the manpage reads:

max xmit (G)
This option controls the maximum packet size that will be nego-
tiated by Samba. The default in Samba 2.2.6 is now 16644
(changed from 65535 in earlier releases) which matches Windows
2000. This allows better performance with Windows NT clients.
The maximum is 65535. In some cases you may find you get better
performance with a smaller value. A value below 2048 is likely
to cause problems.

Default: max xmit = 16644

Example: max xmit = 8192

The next parameter "dead time" is actually used as "deadtime" per the man page, although it could just be depreciated.

Can I simply reload the conf and not have to send a SIGHUP to realize the new values?

Thanks for your input.

JC

I've found in real world testing the option that is directly related with the MTU and window size is max xmit. This option sets the largest block of data Samba will try to write at any one time.

Because the percentage of each block required for overhead falls as the blocks get larger, max xmit is conventionally set as large as possible. It defaults to the protocol's upper limit, which is 64 kilobytes. The smallest value that doesn't cause significant slowdowns is 2048. If it is set low enough, it will limit the largest packet size that Samba will be able to negotiate. The new value of 16644 may work better but as with any tweak your milage may varry.

I believe deadtime vs dead time is depricated like you said.

You sould be able to reload the conf.

Glad that we got to the bottom of the issue! I was worried it was going to be more complex but I'm glad we got it hammered out.

I will still...
 

Continue to play with it. Even though the tweaks worked to improve the transfer time, it by no means meets the performance level of the Win2k machines.

I appreciate all your input. I'll keep this tread posted if I have an epiphany.

JC

i'll continue to hack it with you if you'd like.

can you set the options I posted in the previous post? try it with the new limit and the old limit.

how are you timing the transfers from the win2k box? is it possible to install cygwin and then time a cp?

No reason to.
 

It's painfully clear that the Win2k boxes WAY outperform the Mac box. I'd be willing to bet that the transfer speed is 10 times faster.

JC

Perl script to help in diagnosis
 

Here is a Perl script which might help in diagnosing the problem. You use it like the 'cp' command, specifying a source and destination file. It does the copy and prints out timing info.

To use it with the file you had tried earlier, try running it like this:

testfilecopy /Volumes/company/Technology/ffbgalmac.zip /dev/null

where I specified /dev/null as the destination so nothing would be written on the local disk. This factors out any possible issues with slowness of the local disk.

Cool
 

137.00 seconds: read 19692336 bytes
137.00 seconds: files are closed
Transfer rate: 140.4 KB/s

binmode
 

By the way, if you have Perl installed on your Windows machines, you could run the above 'testfilecopy' script there to get comparable numbers.

But you would have to add two additional lines to the script to make it work on Windows. You need to add 'binmode' commands for both the SRC & DEST filehandles in order to avoid problems with Windows end-of-line handling. So add the following lines after the lines that 'open' the files:

These lines are innocuous on UNIX (e.g. OS X) and so you could add them into the script everywhere if you like.

more of the same from a newbie
 

i am having what i think is a similar problem to this, but i am a relative newbie to mac's, most of the stuff here i am unfamiliar with but this is my situation

12" powerbook OsX 10.2.6 10/100 ethernet, connected as in the Mac OSX hacks book "one wire network", with a CAT-6 cable into my XP pc also with 10/100 ethernet

copying a large file from pc to mac by just dragging it from the pc's shared folder is not too bad, not particularly quick though, but copying a file from mac to pc is incredibly slow

if, as it sounds you have managed to sort this, i would really appreciate if you could help, bearing in mind i am no expert

How are you sharing?
 

Are you using the Windows File sharing option in OS-X, and are you using XP-Pro?

JC

yes i've turned on the windows sharing on OSX and i'm using XP Pro

From what machine?
 

Are you working on the Mac, trasferring files TO and FROM the PC? If so, I doubt that I can help you, since you cannot "tweak" windows file sharing too much that I'm aware of, unless you use third party tools like http://www.tweakinternals.com/. Make sure ALL your updates to XP have been done...M$ bugs...:rolleyes:

If you are working on the PC, and trasferring files TO and FROM the Mac, you will have to dig into the Mac's Samba configuration files, and tweak them. Use some of the info from previous posts in this thread, and go to http://www.samba.org/ for more detailed information about the smb.conf file and it's configuration parameters.

I also found a good article at http://www.opensource.apple.com/proj...l/osxsmb.html.

That should give you a good start.

JC

other thread
 

For the benefit of anyone who comes upon this thread in future, I note that there is another thread which discusses a similar problem only with AFP shares instead of SMB shares.
http://forums.macosxhints.com/showth...=&postid=84058

And I have posted an improved version of my Perl script in that thread.

same sluggishness, solutions may be over my head
 

Hello,
I have a similar sluggish problem:
I and 6 other designers are on OSX.3.8, and connect to a fileserver running windows 2000 server. I have a pc and a mac in front of me. the pc with xp pro connects with no problem, but when I try and connect from the mac, it is painfully slow. It takes forever just to display the contents of a folder (especially if there are a lot of items in that folder), let alone uploading or downloading files. At one point while I was waiting for a folder to display, I got on the pc and surfed 4 folders deeper than the one that was trying to render, copied a file, pasted it on a web server and tested it before the other folder even displayed. Unfortunately(or fortunately for me) I'm the only designer who has the option of using the pc. I ended up posting a file on my mac's web server, and my coworker across campus downloaded a 100MB file in the blink of an eye. Why is the speed that fast over the internet and not to the next building over? Is it a mac issue or should I have the sys admin of the fileserver look into it on his end. Everything that has been said so far in this thread is way over my head. Is there a pseudo-layman's version of a fix?

Networking problems are always hard. There are so many variables - the machines at each end, the software each is running, the software used for doing the file transfer, the cabling, the routers etc that intermediate between the machines, ...
There is usually no quick fix.
You need someone who can look at the whole picture.

One thing you could do is try to get a better diagnosis of the problem by changing some of the variables. E.g. see what the transfer speed is if you use a command-line tool instead of Finder, see what happens if you mount the disk inside of Virtual PC on the Mac, see what happens if you use an FTP client to download files from the server, ...

smb
 

Thanks for the response, Hayne! I went ahead and tried entering smb://<fileserver IP address> in the connect to server dialog, and have gotten significantly better speeds for rendering and scrolling through the directories, but there is some strangeness like little boxes with exes in them replacing some characters. Is there anything I should be aware of connecting that way? I'm going to start down the ugly road of finding all the settings in the big picture over here. (it's a bit like the blind leading the blind in my department)

S

It sounds like there are some characters in the filenames that are not supported by SMB.
You should be aware that SMB does not support all of the metadata that indicates what application is associated with a file. This is likely to cause problems if the files don't have suffixes that indicate their type.

thanks
 

for that info! I'll let my designers know to alwyas include the suffixes. That could have gotten ugly.

S

Hello.
Taking this thread further.

I have the same problem and the same setting as mentioned.
I have a Mandrake 10.1 server with a Linux 2.6.11 custom kernel, samba 3.0.7.
Currently I'm on a Powerbook with wireless and connecting to Mandrake with samba shares. I'm in need of samba because I also have Windows based computers.

Uploading small files to the server takes seconds, large files takes minutes.
Downloading small files from the server takes seconds, large files takes ages.

I've trimmed my smb.conf settings to this :
I've tried to trim my config file. But It's not optimal yet. Some good advices had been great :-D

FROM W2k3 server TO mac os x ... super slow
 

I agree with Hayne that the entire concept needs a global approach. I am frustrated with the networking of Macs and Windows Server 2003. Before, when running W2K server, I had no problem downloading and uploading files between Macs and the server. But since the upgrade to Server 2003, the Macs download files at a snails pace.

I've modified the HOSTS file, setup DNS correctly in Network Pane, created a script that resolves unicast DNS for Macs residing on a Windows ".local" domain...MS UAM, Correct File Sharing procedures, etc...
This is a networking issue, probably concerning how W2K3 Server sets security which affects Macs somehow...

I would like to see more information on what's necessary to enable/disable as a security feature regarding AFP or SMB on the server itself so that Macs can get better performance.


any thoughts?

-- pheidippides --

We dealt with this issue for a long time, and i fiddled with setting even sifted through tcp dump logs for mss setting re-negoiation. I believe the first thing we did was turn up the max xmit. Perhaps when we tested last there were other problems with the os x client however we just re-visited this today and lowered the max xmit to 16644 and bang finder operations are lightning fast now.

[global]
log level = 1
log file = /var/log/samba/log.%m
max log size = 50
socket options = SO_SNDBUF=8760 SO_KEEPALIVE TCP_NODELAY IPTOS_LOWDELAY
max xmit = 16644
read raw = yes
write raw = yes
oplocks = yes
dead time = 15
getwd cache = yes
os level = 255
domain master = yes
preferred master = yes
local master = yes

I've struggling with OSX and SBS 2003 for weeks.

I'm new to OSX, is the global config above on the OSX client side? If so, which directory is this at?

Thanks.

max xmit change
 

ok, I found where the config should go but didn't help. Still as slow as hell.

I've put max xmit = 16664 statement on the smb.conf file, rebooted but no effect.

Any other suggestions? This is killing me.