GTCs
Donate
Couple days we reached my 100% goal for the donation drive (1.500€). The EVE Online community is really the best and I’ll try to do my best to provide a useful and stable service. In the mean time I’ve already ordered my replacement drives, ssds, etc and even opted to get more RAM (all ~2.500€). More memory is always better, ZFS loves more memory and it provides me more flexibility when playing around with other virtual servers.
In the beginning I’ve asked for opinions regarding the storage solution. At this point I was about 90% sure that I’ll go with the hardware raid controller (which I had experience with) and some SAS drives. But you all and some friends poked me in my back and pointed me into a different direction. ZFS! I’ve had used ZFS at work multiple times but always in combination with Solaris. ZFSonLinux has left a bad taste in my mind due to its first linux introduction zfs-fuse) and the license issue (the reasons why zfs is not part of the official kernel). The more I was reading and researching about performance tuning, integrity and features the more I felt comfortable with it. And not having to buy the 600€ HW Raid Controller offered me the possibility to buy more SSDs which I could use as ZIL/SLOG (write-log for sync) and L2ARC (Read Cache) to speed up the overall performance of the storage.
As mentioned yesterday on twitter: I got everything prepared and within 1.5 hours I could swap out the drives, insert the controller, do the cabling and do some basic tests. The migration of the storage itself is/was something that I wanted to do later after I’ve tested everything. Apart from some minor issues (not all of the ram got recognized (again, more on this later) and forgot to check the system clock) it all went smooth.
For the hardware geeks here’s want I ended up with:
- 4x 1TB SAS drives (2 Vendors, 2x WD1002FBYS, 2x ST1000NM0023) with raidz2 (raid6)
- 1x IBM ServeRaid M1015 (LSI2008 based Controller with IT Firmware)
- 4x 100GB Intel DC S3700 SSD (2x for ZIL/SLOG write log, 2x for L2ARC Cache)
- 2x Kingston 30GB SSD (to seperate KVM/Hypervisor and Storage)
- 96GB DDR3 RAM (6×16 max what the mainboard currently support atm)
As mentioned earlier it’s a bit frustrating: Before the upgrade I had 32GB RAM (4×8) inserted and only 24 GB were usable cause one module wasn’t detected. After I replaced the modules with a full set of 6x16GB mods the failure came up again sometimes 1 sometimes 2 modules weren’t correctly detected. This only tells me that there might be an issue (which I already had back in 2011) with the mainboard dimm slot since I own it. It’s still enough RAM to go, play with and make the databases and ZFS happy, but It don’t feel right. At least it’s nothing you’ve to worry about, it’s just me grumble about it 🙂
So what is left to do? Migrate data!
While I’ve already done some tests and migrated some smaller test and dev servers over to the their new home, the main part is still missing: the evemaps server. I’ll likely wait some time (12h-24h) to get get a baseline on how the other stuff is behaving and track how things like IOPS and disk latency are getting affected when I move evemaps over (which be another downtime for 30 minutes to move it over to the new drives).
I just can’t be more then happy to be backed by such a cool community! Thank you! See you in Iceland in april/may for #evefanfest 2014 🙂
