Poor bandwidth (both in general and distributed) [message #43816] |
Tue, 18 October 2011 06:03 |
bwqx
Messages: 2 Registered: October 2011
|
Junior Member |
|
|
kernel: 2.6.32-042stab037.1
tools: default version that came with kernel install
With a clean install of CentOS 5.6, my server is able to do at least 90 MB/s full duplex (I've tested this extensively too, not just assuming this). After installing the openvz kernel from the repo and changing the systcl I tested the bandwith again. I was only able to test the downstream at that time but I still got the same great speeds. Next I created some containers (Debian 6.0 minimal) and I ran the same tests there. One container got to 79 MB/s download and then dropped to 70 MB/s. Another container got 76 MB/s and stayed pretty stable.
Next I tested the upload speed and that was pretty horrible, only getting like 20-30 MB/s and barely 40 MB/s with 1 container and the head node downloading the same thing at once.
Is there anything I can do to make this better? These containers are all using max settings from the "vzsplit -n 5" program. At the moment there's just 2 containers running, but I'm planning on extending to 5 containers. I have already checked the user_beancounters and it's just a clean record for the failcnts there, 0 on everything.
I chose OpenVZ over Xen because I wanted the best possible performance so to say I'm disappointed is putting it mildly..
In short again:
host node without containers capable of 90+ MB/s full duplex, after creating and running 2 containers with max-settings, only changing disk space, they are struggling to get over 30-40 MB/s upload speed. Host node has 24GB RAM, dual quad-core i7 xeon w3520 and 2x2TB RAM in RAID0 (softraid).
[Updated on: Tue, 18 October 2011 06:35] Report message to a moderator
|
|
|
|