[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: ISPmail Lenny tutorial ready

Christoph Haas wrote:
> We are doing MySQL replication, too. The binlog stream takes up ~2 Mbps
> and it takes rather fast servers to keep up with the disk IO load. So I
> can hardly imagine what kind of hardware would be needed to replicate
> 100 Mbps of SQL stream as mentioned earlier. Besides how many changes
> would you need to do per second to get SQL statements making up 10
> MB/second?

The individual servers in my slave pool are certainly not capable of
doing 100 meg each (unless it was blobs or something). The load balancer
is set to give each slave up to 1200 active connections distributing
with a 'least connections' method. To add capacity just add to the
cluster and the load balancer deals with doling it out. It works great,
it's fault tolerant, completely transparent, and scales beyond what a
single server can handle. I hate to admit it with the impressive
hardware being thrown around, but I'm using cheap 1U Celerons with two
non-raid SATA drives (one for OS+logfiles and one for tablespace) behind
the load balancer. ;)

I don't have much going on with binlog updates because it's mostly reads
except for when a customer updates a setting, and that doesn't happen
constantly nor take up much traffic. The write traffic is held on the
slaves in a local-only table as a queue until the daemon that picks it
up for processing passes through and deletes it.


Reply to: