VMware Cloud Community
DaveUSA
Contributor
Contributor

VMWare + Memcached = poor performance

VMWare + Memcached = poor performance.

I did a benchmark between VMWare, OpenVZ (Proxmox) and desktop computer with Memcached installed.

"Memcached is an in-memory key-value store for small chunks of arbitrary data (strings, objects) from results of database calls, API calls, or page rendering."

It uses RAM memory not HDD.

I don't know how to resolve this issue but performance is unacceptable. I was trying different Linux distributions, different kernel versions, different network drivers and nothing has changed.

Summary (requests per second to Memcached instance):

-------------------

VMWare: 1 213 req/s

OpenVZ: 24 109 req/s

Desktop: 36 221 req/s

Benchmark source (c++): http://pastebin.com/aNCe6hpe

BENCHMARK:

========================

localhost - connection via unix socket

remote - remote connection to machine (TCP/IPv4) on the same Node (the same physical server).

VMWare:

-------------------

Connection to localhost:11211: ~ 8 214 req/s

Connection to remote:11211: ~ 1 213 req/s

After turning Memory Affinity setting to ON:

Connection to localhost:11211: ~ 36 512 req/s (rised up)

Connection to remote:11211:  1 213 req/s (without changes)

OpenVZ (much worse machine but performance better than VMWare):

-------------------

Connection to localhost:11211: ~ 26 338 req/s

Connection to remote:11211: ~ 24 109 req/s

Desktop computer:

-------------------

Connection to localhost:11211: ~ 70 023 req/s

Connection to remote:11211: ~ 36 221 req/s

TECHNICAL SPEC:

=========================

Machine used as a base for VMWare 5.0 installation:

-------------------

Fujitsu RX300 S6

CPU: 2 x Intel Xeon X5650 2.67GHz 6 core per processor

RAM: 98GB 1333 MHz

Disk Array:  Eternus DX80/90 S2

HDD: ST3300657SS

Machine used as a base for OpenVZ (Proxmox VE 2.3.13) installation:

-------------------

CPU: 2 x procesor Xeon 4-core E5506 2,13GHz 4MB 4,8 GT/s (Nehalem)

RAM: 12 GB ECC Registered DDR3 RAM 2 Rank ATP (6 x 2048 MB)

Desktop computer:

-------------------

CPU: Intel(R) Core(TM) i5-2500 CPU @ 3.30GHz

RAM: 10GB 1333 MHz

0 Kudos
3 Replies
Linjo
Leadership
Leadership

How is your virtualmachine configured?

Are you running the latest versions of vSphere and VMware Tools?

// Linjo

Best regards, Linjo Please follow me on twitter: @viewgeek If you find this information useful, please award points for "correct" or "helpful".
0 Kudos
vmMarkA
VMware Employee
VMware Employee

A couple process and technical suggestions:

- Comparisons require "exact" same hardware between tests.  I'd suggets picking one server and running the tests on it just changing hypvervisor.

- More detail on the VM config is required: vCPU count, configured memory, etc.

- Check the Power policy on vSphere.  For benchmarking typically one turns of power management in the BIOS or uses High Performance in vSphere.

This article might also provide some assistance:

http://blogs.vmware.com/vsphere/2013/01/troubleshooting-performance-comparisons-with-cheat-sheet.htm...

Cheers

Technical Marketing, Performance Specialist, VCDX @vmMarkA, virtualizationeh.ca
0 Kudos
ziolek
Contributor
Contributor

Hi,

> More detail on the VM config is required: vCPU count, configured memory, etc.

4vCPU, 16GB RAM,

> - Check the Power policy on vSphere.  For benchmarking typically one turns of power management in the BIOS or uses High Performance in vSphere.

We did it.

We tried most of http://www.vmware.com/pdf/Perf_Best_Practices_vSphere5.0.pdf including NUMA afinity, vmx3 network interfaces, etc.

Without effects.

Servers are bored. Physical CPU and RAM aren't overcommit.

We have no idea what I can do else.

Regards,

TZL

0 Kudos