SI is running on the same equipment in KC that it ran on in Seattle. And the bandwidth utilization isn't as high as I'd feared. Still, it can come down quite a bit through efficiencies in the interface.
We literally boxed up all of the existing hardware, flew it to KC, unboxed it, and set it up identically in our ISP.
I'm noticing that the lag seems to be anytime an insert (posting a message) is being done. And a smaller amount of lag on reads, which I suspect may be because of the load caused by inserts.
Because of the kinds of lags I'm seeing, they're most likely a database issue, so we're going to have to suffer through it until the database conversion is done. It's entirely possible and even likely that moving everything to a much more powerful database server, and one whose queries and indexes I can tweak as needed, will completely solve the lag problems, so it's not a good idea for me to try to quickly get up the Oracle learning curve and try to address the lag issues right now. Better to wait until the conversion is done and make those tweaks to a system I already know very well.
I have no idea how to compare Sun equipment with INTC-based equipment, so don't know if this is apples and oranges, but the message database server is running 4 400Mhz processors and talking to an array of 12 10-gig, 10k rpm drives in a RAID5 configuration. If I remember correctly, it has about 2 gig of memory in it.
The new box (Dell 4600) has 2 2.8Ghz Xeon's in it, 8 gig of DDR memory, and a RAID5 setup of 3 36.7-gig 15k rpm drives.
[edit] At face value, I'm looking at 5.6Ghz of computing horsepower (vs 1.6 Ghz), 50% faster hard drives (maybe more because of fewer drives?), and 4 times as much memory, so my uneducated guess is that the new database server will be much more than twice as fast. Maybe as much as 4 times as fast. But though I know nothing about Solaris, I suspect it's a LOT more efficient than Windoze. I suspect there's a "Windoze Penalty" in this setup, but no idea how much it might be. Still, am I correct in thinking this new backend will be about twice as fast? [/edit]
Can anyone who's familiar with both give me an idea how much faster the new 4600 will be? It's my hope that the one box will handily replace the 3 current database servers. 2 of those database servers represent a relatively small workload. It's the message database server that's really having to work its butt off.
For comparison's sake, when iHub's database server was a 2-gig memory, 1.9Ghz machine with a single 120-gig IDE drive (software mirrored onto another identical drive), it was typically running at about 40% CPU utilization and spiking to 90%+ anytime someone would do a search. Since replacing it with a 4600 as described above (but with 4 gig of memory instead of 8), it's usually running at about 2% utilization during the day with occasional "spikes" to 6% on searches. |
