With static/in memory pages serving from a cache or fast server, it's clear the bottleneck is the NIC.

If one gets two nodes and load balances them, are you guaranteed they are on a different physical server, and thus have a higher potential network bandwidth on the physical NIC?

It would seem in that case to be better than a single larger Linode?

I would imagine the performance latency would come the most from CPU processing of the web server and threading.

Remember, if you're using say, Apache + lots of libraries, each page load requires it to load all of those libraries, especially something such as libapache2-mod-php.

If the problem is network-related there's little you could do about it anyway, even with dedicated hardware. You would see more signficant gains in this case through CDNs such as akamai or through geo-location and pointing users to a closer server.

Moving to another linode for this reason would be of insignificant value unless you're CPU/RAM/Disk-bound.

Thanks for response, but...

As I say, in my particular case I can see my 512 linode coping fine in terms of CPU, memory, and disk IO. I'm only serving the initial generated page from Apache - and I'm certain the cache is being hit subsequently.

What's limiting my pages served from memory (via Varnish) is definitely network IO.

So my question was, how do I get greater IO? Does a second linode ensure you're on a separate physical box?

Also, IIRC there is a 50mbit/s cap artificially imposed on individual linodes that they will bump up if you can justify it.

Doesn't need to be on different host to get greater network IO. Each node is guaranteed certain bandwidth, reserved out of total bandwidth of the host. Even if you owned 10 nodes on single host (unlikely but let's assume for the sake of the argument), loadbalanced, that's 10 x the guaranteed bandwidth of single node.

IMHO more smaller nodes is always better than one bigger (unless there are other factors that require bigger disk space, memory) because you then have redundancy.

(Edit: I mean bandwidth, not monthly traffic which is capped much lower than max default sustained bandwidth can cause)