[resending -- left of list addr last time] >> These are clearly over the 300,000K limit -- and the swap stat files >> are less than 1MB. > > Um, you mean the 300 GB limit configured. 307,200,000 KB to be precise. Yes, right. > Which indicates that something other than Squid data cache is going into > those disks. OR, the unlinkd eraser process is not working. Nothing else is going in there. 2000 files which I can visually inspect are of the format /XX/YY/ZZZZZZZZ > You may want to locate and remove the "none" files created by the above > logging. If they are in /squid/cache that is likely your problem. I don't have any 'none' files. The squid 3.1.8 docs says 'none' is the default as well. And lsof shows no files named 'none' opened. > cache_log contains the critical and important administrative messages when > things go horribly wrong. There may be something logged there that will make > your problem clearer. Yes, I'll turn that on. Current statistics are: Store Directory Statistics: Store Entries : 2052 Maximum Swap Size : 204800000 KB Current Store Swap Size: 117010612 KB Current Capacity : 57% used, 43% free Store Directory #0 (aufs): /build/mts/squid/cache FS Block Size 4096 Bytes First level subdirectories: 64 Second level subdirectories: 256 Maximum Size: 204800000 KB Current Size: 117010612 KB Percent Used: 57.13% Filemap bits in use: 2025 of 16384 (12%) Filesystem Space in use: 178800012/422289984 KB (42%) Filesystem Inodes in use: 119414/422421056 (0%) Flags: SELECTED Number of files matches what is actually there: $ sudo find . -type f | wc -l 2026 And summing the ls -s output of all the files gets me 175,731,252 $ sudo du -sk /build/mts/squid/ 175740120 /build/mts/squid/ Which matches the du. I suspect something is wrong in the accounting code. Possibly somewhere a signed/unsigned or 32bit/64bit incorrect cast.