LSMCD Tuning


We are running LS Memcached on a very busy form website (peak traffic 16,000 requests per minute).
We have set up 4 dedicated lsmemcached servers (2 CPU, 4 GB RAM) as a replication array.

Can anyone give some guidance on the best node.conf parameters for this setup?

How does Repl.HeartBeatReq and Repl.HeartBeatRetry affect performance?

In the docs, the loopback address is used for:
Repl.ListenSvrAddr, REPL.DispatchAddr, CACHED.PRIADDR & CACHED.ADDR
Should this be changed to the servers private IP address?
I have seen Repl.LbAddrs with both the loopback address and the host IP in the replication array. Which is correct?

Cached.MemMaxSz: What is the correct setting for a high traffic site with 4 lsmemcached servers?
I understand the default is 64000000 (MB). Does increasing this value help with performance?

Finally, what is the correct way to assign Cached.Slice.Priority.0-8 among the 4 servers?

Thank you for any guidance you can provide.
I have searched the docs, but cannot find any use case scenarios.

Attached is our existing node.conf for the first of the 4 servers


Sorry, let me try again (hit reply too soon):
Hi Mr. Wizard!
(Cool handle!).
Let me address your questions:
- Repl.HeartBeatReq and Repl.HeartBeatRetry are not used.
- Repl.LbAddrs is supposed to contain a list of all servers in the replication array. Because they're usually separate machines (and that sounds like your configuration) you'd use the real IP address.
- Repl.ListenSvrAddr must be a member of LbAddrs so use the real IP address.
- Repl.DispatchAddr should be the local address.
- Repl.PriAddr is not used.
- Cached.Addr is the public address and is required. I'd use the real address in your environment.
Check out: for details.
- We don't have any experience in setting priority across members.

Cached.MemMaxSz is quite important as it is the total amount of memory to be stored in the shared memory hash table before data is taken out. Since hits are important, you want it to be large enough to hold whatever data is likely to be relevant. However, it can't exceed the legal amount of shared memory in your system, and shouldn't be larger than the amount of data that you're likely to store there. So much depends on your application's use of the cache, we can't suggest much more than ask your application's vendor.

We don't have a lot of customers using replication so we don't have a lot of experience in optimizing it.
I was not able to see your node.txt file. Can you send it to my email address:

Bob Perper