memcached is not effective

[Balakumar]

We are running IP.Board on 2 nodes, and we have conducted a load test with both read and write operations.

But the memcached is using only 23 MB, and also found that the application is highly CPU intensive.

Is there anyway we can optimize this, can anyone please give suggestions on this?

Thanks,

Bala

[Grumpy]

Can you post more information? Like a LOT more information?

[Balakumar]

Base : Xenserver

Server Details :

2 APP VMs : 4GB RAM, 8 core VCPU, 50GB Disk Space (resides on same Xenserver)

1 DB VM : 4GB RAM, 4 Core VCPU, 50GB Disk Space (resides on different Xenserver from app nodes)

We ran the load test using JMeter and it is about 1700 read users doing 5 different page reads, 20 write users doing new posts in about 15 mins time total.

With the above load test we are seeing the mecached been used only 23MB in total and a lot of DB requests being made compared to memcached, ideally all the 5 different read pages should mainly be cached by memcache and be able to serve from there. Also we have seen the DB CPU usage is way too high. It is about 10% idle CPU on an average of the total run time.

Also we haev seen the application servers also have a high CPU usae and which has an average idle time of 25% with the above load run. The test has almost crashed the VM's.

Thanks,

Bala

[Grumpy]

What makes you say memcache is not effective? What do you mean by "effective"?

Is your DB server SSD based? 5.4k, 7.2k, 10, 15k HD?

What is the process model for your cores?

When using JMeter, did you have a step up process or was it for consistent?

What is the hit rate/miss rate on you memcache? IPB does not use page level caching, as such, you are expected to still see DB calls even with memcache usage.

Do you have a large number of resource which would warrant a usage greater than 23MB on memcached?

Lastly, IPB does have high CPU usage (relative to the rest of the resources), this is a fact that you can't really do much about beyond optimizing php with apc/xcache and such.