We use some Samba servers running SuSE 7.1 and samba 2.2.1a. These servers are sometimes very slow while delivering data over the network. It is like a short sleep-mode followed by a wake up. We use Win 95, 98 and 98SE Clients. Any ideas? Thanks from germany, Andre Klocke -------------- next part -------------- HTML attachment scrubbed and removed
On Friday 26 October 2001 08:02, Andre Klocke wrote:> We use some Samba servers running SuSE 7.1 and samba 2.2.1a. > > These servers are sometimes very slow while delivering data over the > network. It is like a short sleep-mode followed by a wake up. We use Win > 95, 98 and 98SE Clients.You're going to have to do a little more diagnostic work. Run "top" or a similar program while the problem is occurring to see what the machine is doing ... check smbstatus to see what kind of memory usage is occurring. Is the swap file being used a lot? How much RAM do you have vs. how many connections are you using? Are there any errors in the log file when the machine is slow? Have you tried increasing the log leve to see if something shows up. I assume that you have checked to make sure that nothing on the server is going into power saving mode ever, but double-check. What speed HDDs do you have? You may just be hitting a situation where your server hardware is being worked hard and is having trouble keeping up. Until you do a little more reseach you won't know for sure. -- Bill Moran Potential Technology technical services http://www.potentialtech.com
We're trying to use Samba here to share out data from our Unix NAS server to
PC (Win2K) clients. However, we're experiencing a problem in that saving a
project from one of the applications (Petrel, used for subsurface analysis
of oil/gas wells) can take 10 seconds, whereas saving to a win2k server
takes ~2 seconds. Obviously, something is not performing...
After a lot of poring over truss files, I've discovered two causes for
concern:
1. there are 5394 stat64() calls during the saving of the project, taking in
total >5s
2. there are 43209 getdents64() calls during the saving, also taking in
total >5s; most of these getdents calls follow an open() of the diretory
holding the files. This open() happens 1053 times (i.e. ~ 2 x the number of
files)
(for reference, the above timings are based on using truss -D -p <pid> -o
<filename> and then running awk '{ s+=$1 } END { print s }' to get
a total)
Note that the figures are somewhat prone to inflation as using truss on the
process slows things down.
This would appear to be the bottleneck in the system, but I'm not sure what
I can do about it.
Some more information about the data: the application stores 526 files in
one directory, each with a name like:
affbae41-0628-42a5-b053-8712a7673567.dat
Other projects store a different number of files. Total disk space used is
~200MB, with file sizes ranging from a few hundred bytes up to 14MB.
The name seems rather random, but follows the same pattern of alphanumerics
and hyphens.
I've tried Samba on AIX (on a p660), Solaris (Ultra 1, v880 & Sunblade)
and
IRIX (12 CPU server), all with similar results (i.e. it takes significantly
longer on Samba than on the Windows server). All are using local disk &
native filesystem (i.e. JFS, UFS or XFS as appropriate).
Samba version is 2.2.2, but I've tried 2.2.6 and it gives the same result.
Does anyone have any idea on how to improve things?
Sorry for the length of the post, but I wanted to include as much
information as possible from the start.
John Riddoch
Unix Project Engineer
Shell Information Technology International Limited
Loirston House, Wellington Road, Altens, Aberdeen AB12 3BH
Tel: +44(0)1224 21 7660 Fax: 7502
Email: John.E.Riddoch@is.shell.com
Internet: http://www.shell.com
Hello, I have a severe performance problem when connecting a M$-Client PC with a Bootdisk and TCP/IP-Stack to my Samba server. Both machines are linked via a 100Mbit crosslink connection. The server is running a SuSE Linux 7.2 with Kernel 2.4.19, the smb-fs and the NIC is compiled into the kernel. The Client uses a DOS 6.2 bootdisk with the M$ TCPIP-stack version 1.02. All M$ drivers load fine, the 'net use' command works too. Then I start Symantecs 'ghost' to create an image from the clients HD and put it to the used Samba share on the server. The Performance is poooor (<1Mbyte/min), tests with a W2K server on the same hardware showed better values (>200Mbyte/min) :-( . What can I do to improve this poor samba throughput??? Regards Ferdi -- +++ GMX - Mail, Messaging & more http://www.gmx.net +++ NEU: Mit GMX ins Internet. Rund um die Uhr f?r 1 ct/ Min. surfen!
We currently are running Samba 2.2.2 on a Sun E3000 under Solaris 8 and we're seeing very puzzling performance issues with a visualization package called VisView Professional. A model is made up of a number of sub-assemblies. When a user opens a model that is stored on a Windows server, it loads and displays in about two minutes. If the same model is opened from my Samba server, it takes over eleven minutes. The time required to copy the directories that contain the sub-assemblies from either server to the local hard drive is between six minutes and six and one half minutes, so I don't think there is an issue with Samba itself. Doing a tcpdump shows a number of "file not found" messages that may have been part of the model, but are no longer needed, so no longer exist (I'm guessing that's the reason). We're trying to keep the users from keeping two copies of the data. Any Suggestions? Mike