-
HyperFS SAN vs. StorNext ?
Anybody used Falconstor’s HyperFS SAN software yet?
I got a demo of the SAN software and want to get an idea of what to expect as far as performance in multi-stream reads and writes as well as single stream benchmarks.
Also anybody running StorNext with similar specs? What benchmarks are you getting for comparison?
My environment is only HD, no film. I have 4 Final Cut Edit bays for HD and a 4 windows servers including a DVS Clipster for DPX. Only 2 edit bays are expected to be editing at the same time as Clipster is operating. So pretty much 3-4 streams simultaneously.
Specs:
2x 8Gb Qlogic 5800 series switches
2 MDC’s running RedHat and using dual port QLA2560 8Gb hbas
4 Windows clients using Atto 8Gb dual port cards
4 Mac 10.6 clients using Atto 8Gb dual port cardsNo MP right now as I want to get good single port marks.
Hitachi AMS-2300v2 dual controller storage with 48 drive SATA arrays and SAS metadata luns. 8 ports to the switch, 2 zoned for metadata only and the other 6 for data.
Luns are:
4 data luns = 10D+2P with 2TB SATA drives , 256K chunk
The 4 luns are balanced across 2 controllers and each has its own 8Gb port.
1 metalun = mirrored 15K sas drives, 64K chunkI set up 1 volume group with HyperFS with four 17TB luns in a stripe using 1MB stripe and 1 MB allocation.
Preliminary test are showing odd results for Mac OSX. Using AJA’s notorious system test to get a quicky reading. Controller write cache enabled on metadata servers.
1Gb file size, 1920×1080 10-bit framesize
Windows is getting me 484 W / 376 Read.
Mac is getting me 360 W / 290 Read. -
4 Replies
-
Not that this will necessarily change your results, but I would suggest using a larger file size in your AJA tests, say 8GB or 16GB to try to make sure that test file puts a more constant hit to your storage.
-Caspian
Product Specialist
Studio Network Solutions -
The Aja test is not a good representation of FCP traffic. It uses a smaller IO size and it’s not Async IO either. FCP does larger IOs in parallel.
Steve Modica
CTO, Small Tree Communications -
Yeah,
I switched to using the compiled for mac version of lmdd.
I also used Brite systems benchmark, which is awesome, btw.I was pinging it pretty hard with iometer as well from the windows side.
HyperFS seems to fly with a Windows environment but I’m running into performance issues on the Mac client. Its still fast, but significantly less than a linux or Windows box on the san.
-
John, what that that “Brite” benchmark, and where can I get it?
Brian
Reply to this Discussion! Login or Sign Up