Tiger Server 10.4.11, AFP and SMB performance grinds to a halt.
Posted on 2008-10-08
XServe, 2 x 2.66GHz Dual-Core Intel Xeon with 4GB RAM;
System drive is internal SATA RAID1 68GB, 58GB free;
XServe RAID (12 x 750GB, RAID50, 6.82TB single volume) directly attached, PTP, 2GBps;
Services are AFP, NFS, SMB and Open Directory Replica (Kerberised);
Only one share point is defined;
Connected via one built-in Ethernet port to:
1 x Gigabit switch (3com 2948 SFP Plus);
Clients: 8 Mac Pros, running Leopard 10.5.4 and/or XP 32 or VIsta 64, connected via CAT6 tie-lines, with CAT5e patch cables.
At least once a day, I have to reboot the server. AFP winds up to 100% CPU or more, using 203 threads, and the system memory reaches 3.96GB, although 3.13GB of that is inactive. I have caught one instance of SMB doing the same, at the same time as AFP. When this happens, network performance drops from about 70MBps to a few KBps, and all network ports on the server are affected.
So far, I have replaced our original very expensive switches with the one mentioned above;
I performed a clean install of the system;
I tried increasing the Quantum value for WAN to match LAN:
[defaults write -g com.apple.AppleShareClientCore -dict-add afp_wan_quantum -int 131072];
I tested the performance of the XServe RAID locally from the server during one of these slow-downs, and the RAID is performing normally.
The logs are not reporting anything to suggest a problem that may affect network performance that I can see.
Does anyone have any idea where to go from here?