The gateway router gets new software and other things
We have now upgraded the OS on the gateway router that was giving us all the trouble last week. This was done during downtime on Tuesday and since then there have been no reported incidents of packetloss or instability that can be traced to the gateway router. So far so good. Let´s hope it was just the software.
The router type is Cisco 7400 which is supposed to handle a throughput of 300kpps under optimal conditions (unrealistic I suppose) but it was breaking down at 20kpps so something didn´t quite seem right 0_o .
Anyways. We are now giving the cabling and electricity issues a thorough revision. I know some of you have been sceptical of our UPS configuration but it has not been that which has failed, rather the cabling in the cabinets themselves. All the SOL and Proxy servers only have a single power supply while the DB machines have two. Each cabinet then has two separate feeds and the DB is connected to both while SOL and Proxy servers alternate.
So when cabinet wiring fails resulting in a short circuit it usually means that one of two powerfeeds shuts down. This has no effect on the DB which continues to run on the other feed but it usually takes out a few SOL and Proxy servers. When we started out designing the cluster we figured that power failures would be a rare occurence and the risk acceptable since the whole hosting center is on multiple redundant backup systems, both UPS and auxilary fuel driven backup power. This is being revised now since we have had two power failures (attributable to cabinet wiring) in the last 5 months.
Then we are also upgrading the DB harddrives. We have purchased a fiber disk storage server from IBM (FAStT 600) which we plan to install in the next 2-3 weeks. This will increase disk I/O capacity of the DB greatly if everything goes as planned. Disk I/O has been identified as a performance bottleneck during peak load times. Scaling was also an issue in the previous solution. This disk storage solution scales to 16TB which should be sufficient for a while :-) (The DB now is a modest 60GB and grows at a rate of 0.5GB per day.)
The growth factor however has been steadily increasing despite efforts to harvest logs to a separate data warehouse and other efforts to keep the database lean for performance reasons. So I would like to take this oppurtunity to ask all of you to refrain from any acitvity within eve that is likely to increase db size such as trading, figthing pirates, chatting, undocking docking, manufacturing etc. :-D Cheers, RB.