Hi all, I'm having a few problem with a Squid web cache server running RH6.2 with a 2.2.16-3 Kernel The system is a pentium II 350Mhz with 512Mbytes of RAM and 24Gbytes of disk space. It runs squid 2.3 and i also use ipchaines to redirect any requests to port 80 to the web cahe port on 3128. Elsewhere in our network I redirect web requests to port 80 on the above server. The kenel is compile with large socket buffers and I have modified the rmem_default,wmem_default,rmem_max and wmem_max parameters to try and increase the number of socket buffers available. ( As far as I can see, in order to gfet > 32K of buffers you have to set the default parameters to be > 64K ) Anyway, when I do a netstat -s, under the TCPExt section I geet a large number of "packets dropped from out-of-order queue because of socket buffer overrun" and some "packets prunes from receive queue because of socket buffer overrun" So ..... 1). on solaris netstat -a gives you a list of connections plus the buffer sizes negotiated for each connection. On linux, setting these buffer sizes seems to be a black art as I can't find any way of finding out what is used. I *think* I've set them up correctly, but there doesn't seem to be a way of finding out.... unless someone knows otherwise 2). One web page suggested changing the contents of a header file and recompiling the kernel so that the requested buffer size isn't scaled down by 2. Isd this my only option to increase these buffers over the 32K limit? 3). Is there anywhere else I should be looking to get rid of these overrun problems 4). Does the 2.4 kernel have any improvements in this area? Any help appreciated alex ------------------- Alex Sharaz [EMAIL PROTECTED] * This e-mail message was sent with Execmail V5.1 * _______________________________________________ Redhat-devel-list mailing list [EMAIL PROTECTED] https://listman.redhat.com/mailman/listinfo/redhat-devel-list
