Squid reloads on failed assertion

We're running squid-2.3.STABLE1 under linux-2.2.14 w/SMP. I've
compiled squid with async-io enabled (with async-ufs) and with 4096
fds (increased as per Henrik's web page). We use SkyCache/Cidera to
populate the cache. The whole thing runs great, save that every 15-30
minutes it logs the following in cache.log:

2000/03/28 11:38:44| storeAufsOpenDone: (2) No such file or directory
2000/03/28 11:38:44| /usr/local/squid0/05/45/00054555
2000/03/28 11:38:44| assertion failed: store_swapin.c:80: "errflag <= 0"
2000/03/28 11:38:47| Starting Squid Cache version 2.3.STABLE1 for i686-pc-linux-gnu...
2000/03/28 11:38:47| Process ID 8992
2000/03/28 11:38:47| With 4096 file descriptors available
2000/03/28 11:38:47| Performing DNS Tests...
2000/03/28 11:38:47| Successful DNS name lookup tests...
2000/03/28 11:38:47| DNS Socket created on FD 1
2000/03/28 11:38:47| idnsParseResolvConf: nameserver 127.0.0.1
2000/03/28 11:38:47| idnsAddNameserver: Added nameserver #0: 127.0.0.1
2000/03/28 11:38:47| idnsParseResolvConf: nameserver 204.192.96.20
2000/03/28 11:38:47| idnsAddNameserver: Added nameserver #1: 204.192.96.20
2000/03/28 11:38:47| idnsParseResolvConf: nameserver 204.192.96.18
2000/03/28 11:38:47| idnsAddNameserver: Added nameserver #2: 204.192.96.18
2000/03/28 11:38:47| Unlinkd is disabled
2000/03/28 11:38:47| Swap maxSize 16695296 KB, estimated 1284253 objects
2000/03/28 11:38:47| Target number of buckets: 25685
2000/03/28 11:38:47| Using 32768 Store buckets
2000/03/28 11:38:47| Max Mem size: 8192 KB
2000/03/28 11:38:47| Max Swap size: 16695296 KB
2000/03/28 11:38:47| Rebuilding storage in /usr/local/squid0 (DIRTY)
2000/03/28 11:38:47| Rebuilding storage in /usr/local/squid1 (DIRTY)
2000/03/28 11:38:47| Set Current Directory to /usr/local/squid0
2000/03/28 11:38:47| Loaded Icons.
2000/03/28 11:38:47| Accepting HTTP connections at 0.0.0.0, port 3128, FD 19.
2000/03/28 11:38:47| Accepting ICP messages at 0.0.0.0, port 3130, FD 20.
2000/03/28 11:38:47| Outgoing ICP messages on port 3130, FD 21.
2000/03/28 11:38:47| Outgoing WCCP messages on port 2048, FD 23.
2000/03/28 11:38:47| Ready to serve requests.
2000/03/28 11:38:48| Configuring Sibling frognet.downlink.skycache.com/3134/3132
2000/03/28 11:41:39| Done reading /usr/local/squid1 swaplog (720740 entries)
2000/03/28 11:41:46| Done reading /usr/local/squid0 swaplog (756851 entries)
2000/03/28 11:41:46| Finished rebuilding storage from disk.
2000/03/28 11:41:46| 1467830 Entries scanned
2000/03/28 11:41:46| 0 Invalid entries.
2000/03/28 11:41:46| 0 With invalid flags.
2000/03/28 11:41:46| 1458102 Objects loaded.
2000/03/28 11:41:46| 0 Objects expired.
2000/03/28 11:41:46| 9712 Objects cancelled.
2000/03/28 11:41:46| 148 Duplicate URLs purged.
2000/03/28 11:41:46| 0 Swapfile clashes avoided.
2000/03/28 11:41:46| Took 179.4 seconds (8127.2 objects/sec).
2000/03/28 11:41:46| Beginning Validation Procedure
2000/03/28 11:41:52| 262144 Entries Validated so far.
2000/03/28 11:41:58| 524288 Entries Validated so far.
2000/03/28 11:42:03| 786432 Entries Validated so far.
2000/03/28 11:42:09| 1048576 Entries Validated so far.
2000/03/28 11:42:14| 1310720 Entries Validated so far.
2000/03/28 11:42:17| Completed Validation Procedure
2000/03/28 11:42:17| Validated 1457936 Entries
2000/03/28 11:42:17| store_swap_size = 12452645k
2000/03/28 11:42:18| storeLateRelease: released 721 objects

During this time, mean service times triple because of the load
increase from reloading the object database and because it won't
register any hits until the objects are validated.

Should I eliminate the async-io? I've noticed several big warning
signs that say "this might not be stable," but don't want to disable
it unless necessary because it really boosts performance.

Thanks for any input.

    Clay

--
Clay Fouts
Network Administrator
FrogNet, Inc.