I'd like to know so more about the cache manager that you are planning for 5.8. I would like to see something like a cache pool with the ability to set how much total system RAM to use for caching, and be able to set minimum and maximum amounts of cache per target, as well as a low, medium, high priority. Kind of like the resource allocation in VMware. I would also like to be able to set the cache write/read ratios per target to 0/100, 25/75, 50/50, 75/25, 100/0. Our next StarWind boxes that I'll most likely be building in 3-6 months will be 2 HP DL585 G7 Quad 12 Core 2.5GHz with 512GB RAM each, 4 LSI 9285-8E controllers each, a 6 port 10GbE Interface Masters PCI-E x16 each for iSCSI. I am trying to decide on the sync channels. 40GbE would be nice if you have some recommendations on a solution that wouldn't require me adding a switch for it. The storage will be DataON DNS-1660 4U 60 drive dual I/O SAS JBODs. We'll be starting with 2 of the JBODs connected to each server, and have the ability to add 2 more to each server with each JBOD having a dedicated 9285-8E. The drives for mass storage will be Seagate 3TB SAS 7.2K drives, and I'm considering some Cheetah 15K.7 600GB drives and RAID 10's for some of our smaller, virtualized databases. Technically, we could do 4 JBODs cascaded from each 9285-8E, but I think additional servers with more RAM for cache would make more sense.
The other ability I need is to asynchronously replicate data between two StarWind HA SANs. The plan is to move the current SAN to our location in the next city over with a point to point fiber connection between them and the new SAN that would be implemented in our Data Center. I currently have a pair of 10GbE fiber connections trunked to 20GbE between our Corporate office and the Data Center, and will have reconfigured two of our older 24TB storage servers to an HA SAN in about 2 weeks that can be used to test and develop this.
Thanks for all your hard work on this software. I'm more impressed with each new version. I bet you all are making some of the bigger companies more than a little nervous.
anton (staff) wrote:Wow! These are real good numbers! Congratulations. And you guys are not greedy when it comes to buying new hardware I'm jealous.
Yes, we've definitely "did some changes to non-cached I/O" as well
SQLIO is nice and it does indeed pop up numbers SQL Server should hit (sick!) but... Do we have any chance to see Intel I/O Meter and Intel NAS Performance Toolkit test run results as well?
Thank you very much for cooperation!