On Sat, 20 May 2000, Alan Cox wrote:
> Latency has nothing to do with the link speed or very litle. I actually
> suspect 1Gbit versus 100Mbit with an apache server would make no
> difference. You need to get latency down - that means tweaking apache
> and considerign something like a netapp as the nfs server
Well shoot a hole in my testimonial.
We do 10-14Mb to the Internet with 100Mb on the back. NFS works fine for it.
However, I keep Apache well tuned and we use a beefed up NetApp for NFS.
The bandwidth on the back end won't make much difference. In fact, assuming a
decent "disk" cache on the web server, 100Mb should work fine. The latency
comes from the two ends. For each simple NFS operation (like a stat()
call) the client packs and transmits a request, the server unpacks and handles
the request, the server packs and transmits the response, and the client
unpacks the response.
> or for static content swapping the NFS for a replicating content engine
> - eg pushing stuff out with rsync
For static content, rsync would be a good solution. Brent didn't specify, but
I have to assume cgi-based writes are allowed, which makes the scenario more
interesting. If it's mostly static, maybe use NFS or 1 big box and use an
httpd accelerator (i.e. squid) server to take the brunt of the requests.
-- Brian
-
To unsubscribe from this list: send the line "unsubscribe linux-net" in
the body of a message to [EMAIL PROTECTED]
