Home › Forums › OS X Server and Client Discussion › File Serving › AFP Slow! Need help fast!
- This topic has 16 replies, 10 voices, and was last updated 19 years, 6 months ago by
macmanjc.
-
AuthorPosts
-
May 5, 2005 at 4:39 pm #361532Josh HurdParticipant
I have a dual 1.3 GHz G4 xServe running AFP and nothing else. The volume that is being shared is striped across 3 disks.
I am receiving complaints and have confirmed myself that the AFP service becomes unresponsive for up to 20-30 seconds on every transaction. That is mounting the share, descending into directories, copying files, etc.
When this takes place I have about 8-10 users actively using AFP and another 20 or so that are idle. The AFP process is using 98% or more of the CPUs. Data transfer rates over the network never get above 175 K per sec.
The system is otherwise reasonably responsive with this load on AFP. Other services respond ok. Apps launch fine and are usable.
This is becoming a major problem and I really need to find a solution.
Any help greatly appreciated!
Josh
May 5, 2005 at 6:27 pm #361534tboneParticipantTell us what you’ve done to troubleshoot so far. Looked at the AFP log? system log? Applied any updates lately? Rebooted?
Trent
May 6, 2005 at 7:11 pm #361557Josh HurdParticipantNothing unusual in any of the logs that I could find. AFP access and error logs are clean. syslog shows no problems. Rebooting does not seem to make any difference. The system is a production server so rebooting it is not a viable troubleshooting technique most of the time. Rebooting does solve the problem for the latest incident but the next one creeps up all the same. I have advised the users who use this server the most to call me when this occurs. Some days I get no calls other days I get 5-6.
Is there a way to spawn extra AFP processes? Would this help?
Would increasing the max thread for the default 40 help?Thanks for the help!
JoshMay 6, 2005 at 7:21 pm #361558Josh HurdParticipantoh… no recent updates or other changes to this system.
The problem has been persistent for some time now. A recent (2 months ago) rebuild of the entire system; new OS, formatted and reconfigured all drives did not have an effect on this issue.May 6, 2005 at 8:09 pm #361562andrinaParticipantIs the problem all the time also – i.e. at 3AM with only one person connected is it still slow? Do your users leave their machines connected to the share all the time, or do they log out at night? What size files are you working with also?
May 7, 2005 at 10:18 pm #361564Josh HurdParticipantThe system is a dual 1GHz G4 Xserve running Server 10.3.8
This problem only seems evident during ‘crunch time’ when about 8-10 people are hitting it reasonably hard. File sizes are probably 10-50 megs. Again let me stress that network throughput never exceeds about 200 Kb/sec according to Server Admin any way.
Some users stay connected all the the time but the ones who are a victim of this problem generally shut down their machines. I have also tried setting idle disconnects to clean up the connections a bit. To no avail.It did just occur to me that some users may well be editing files directly from the server. I mean opening a file that lives on the server in say PhotoShop as opposed to copying the file to their local machine first. Does anyone have info on how this would affect performance?
Thanks again all!
JoshMay 8, 2005 at 2:34 pm #361567chiefgeekParticipantWhat kind of network switch do you have and what speed? What kind of router and connection to the internet?
We had a similar problem after installing an Asante 24 port GigE switch that wasn’t getting along well with our Cisco T1 router. Moving the router to another switch that is linked to the main switch resolved our issue, which was a problem with auto-negotiation between the two boxes. It was affecting our throughtput on the intranet as well as the internet.
Have you tried running Helios LanTest? It can be quite useful. Look for unbalanced read / write throughput. You can install it on multiple client machines and run concurrently. You need to make sure you set the Target (cmd-T) to a shared directory to test network throughput.
We have the same Xserve and regularly have 5-8 machines reading and writing large image files on the server. The clients range from Win2k Server and Client to G4s running OS9 to G5s running OSX. We work on image files stored on the server without copying them to the clients and generally see throughput rates in LanTest of 30-50 MB / sec over the network.
HTH
May 12, 2005 at 6:22 pm #361658Josh HurdParticipantInteresting.
We have an 2 Asante 10/100 switches and a Cisco 10/100 switch. The LAN router is a m0n0Wall.
I will try shifting things around a bit.
Any idea how the LAN hardware is causing the CPUs in my server to peg out?Thanks,
JoshMay 13, 2005 at 2:37 am #361665chiefgeekParticipantPull the Cisco and give it a try.
Not sure exactly how or even if the lan hw is causing the servers cpus to peg, unless there is congestion / confusion (do you like my tech terms) and it is causing the machines to resend the same data over and over.
Errors with Half / Full duplex on auto-negiotiation do strange things.
May 20, 2005 at 12:42 am #361729MarkParticipantWe too are having the same/similar problem. We have four xserves with 10.3.9 server. Each serves Windows 2000 and Mac OS 10.3.9 clients. At random times the AFP service will slow to a crawl (64K/sec). All other services, FTP, Samba work at full speed (Gig core to 100Mbit desktop). There is nothing of note in the logs, AFP error, system.log, etc. The only fix is to stop the AFP service (therefore disconnecting all the Mac users, preventing them from saving their work) and start it again. The speed then returns to normal. I tried killing lookupd, no change. Tried sharing another folder from the boot drive and copying a file to it, no different. Can you send a signal to the AFP process to restart but not disconnect everyone? Any help would be great.
May 22, 2005 at 4:11 am #361735AMSRParticipantThere were fixes that went into 10.3.9 to help address this. It seems also that creating file structures with hundreds/thousands of files that need to be listed when you hit the directory exacerbates this. As does setting options in the Finder such as “Calculate all folder sizes” etc…
May 23, 2005 at 5:42 pm #361755Josh HurdParticipantThanks for the input! Glad I am not the only one seeing this problem.
The server is question is still running 10.3.8 so an upgrade could be helpful.One thing I found to help spread out the occurrences was increasing the max threads of the AFP server. You can do this with the command:
serveradmin settings maxThreads = xxx
I set mine to 250 which seems to me extreme but I only have to restart the AFP server about once a week or so now.Any more suggestions or ideas greatly appreciated!
Josh
May 23, 2005 at 7:30 pm #361756Josh HurdParticipantBummer… I restarted the AFP server about 2 hours ago and it is already acting up!
Guess my theory on the max threads is blown out of the water…September 2, 2005 at 5:06 pm #363088Justin BurnsParticipant‘lo everyone.
‘m seeing this problem as well on our dual 1.33 g4 xserve, running 10.3.9. We have about 50 OD clients running 10.3.9 and 10.4.2. No AD integration and no windows clients accessing resources on the server. First discovered the AFP process going crazy about a week ago, although I had noticed speed issues for a couple weeks.
Troubleshooting:
Freed up significant disk space (from 93% to 66% full)
Removed all .DS_Store files recursively in /Users sharepoint
Applied the latest round of security patches
Techtool pro intermediate tests (100%)Logging off all users from the server seems to bring the process under control. But even SSH’ing into my workstation at 2am and accessing files on the share was enough to bring AppleFileserver’s CPU usage up to between 40 and 90 percent.
One thing of note is that we just moved from dumb to managed switches (HP Procurve) within the same timetable that this has been happening.
I’m in the process of moving users from networked to mobile home directories, in the hopes of minimizing the impact from this and any other network wonkiness. Hopefully if we all compare notes we’ll find the common denominator here and be able to put an end to it.
-justinb
September 29, 2005 at 10:46 pm #363412dragonmacParticipantWell I’ve been in this boat for some time and there has been no real fix as of yet. I hope to have 10.4.2 on all Servers and Clients soon and Mobile-Home Dir setup when Tiger server is up. I through money at the problem in different areas (new network switches, better HD’s, Outside Consultants) no dice everyone is stumped.
My bottom line is the Finder in 10.3.x< stinks in many cases on network volumes and AFP server in XServer Still needs work. Hopfully 10.4 Server will help.
If I didn’t love Apple I’d hate Mac’s… well at least the Server. ASIP6 worked great and I almost went back. -
AuthorPosts
- You must be logged in to reply to this topic.
Comments are closed