Repeatable ZFS "kmem map too small" panic on 8.0-STABLE

[Gary Corcoran]

Adam McDougall wrote:
> Put this in /boot/loader.conf:
> vm.kmem_size="20G"
> 
> It is intentionally higher than your amount of ram.

Would you mind explaining...
1) why this fixes the kmem_map too small problem ?
2) why it should be larger than the amount of RAM, and by how much ?

Thanks,
Gary

> On 01/21/10 13:05, Doug Poland wrote:
>> Hello,
>>
>> I've got an 8.0-STABLE (amd64) box with 4GB RAM. The machine is running
>> off a 6-disk RAIDZ1 booting from GPT. The box consistently panics on
>> unixbench's fsdisk program.
>>
>> I have been gathering some metrics in an attempt to isolate the
>> parameters that are significant, but I admit I do not really understand
>> all the relationships. At this point, I have nothing set in
>> /boot/loader.conf
>>
>> Here are some of the values I was recording within seconds of the panic:
>>
>> # dmesg | grep memory
>> real memory = 4294967296 (4096 MB)
>> avail memory = 3961372672 (3777 MB)
>>
>> kstat.zfs.misc.arcstats.size: 308522248
>> vfs.zfs.arc_max: 829480960
>> vfs.zfs.arc_meta_limit: 207370240
>> vfs.zfs.arc_meta_used: 165575944
>> vfs.zfs.arc_min: 103685120
>> vm.kmem_size: 1327169536
>> vm.kmem_size_max: 329853485875
>>
>> # vmstat -m | egrep 'InUse|solaris'
>> Type InUse MemUse HighUse
>> solaris 491349 1316172K -
>>
>> % zpool status
>> pool: bethesda
>> state: ONLINE
>> scrub: none requested
>> config:
>>
>> NAME STATE READ WRITE CKSUM
>> bethesda ONLINE 0 0 0
>> raidz1 ONLINE 0 0 0
>> gpt/disk0 ONLINE 0 0 0
>> gpt/disk1 ONLINE 0 0 0
>> gpt/disk2 ONLINE 0 0 0
>> gpt/disk3 ONLINE 0 0 0
>> gpt/disk4 ONLINE 0 0 0
>> gpt/disk5 ONLINE 0 0 0
>>
>> errors: No known data errors
>>
>> My concern is if I can panic the box with a simple file system
>> benchmark, what will happen when I rysnc files across a 1GB LAN
>> connection? I am very willing to run any number of tests and tweek ZFS
>> as necessary. Please advise.
>>