KernSafe Technology

Testing your latest software and noticing strange behaviour when copying large files.

Here is my test setup:

Istorage server: Windows 2008 Web Edition with 12 gig and raid 10 for the storage of image files. Dual 10G Nics (Jumbo Frames enabled). using img files.

Initiator: windows 2008 datacenter with dial Gigabit Nics (Jumbo Frames) using MPIO. Raid 10 on drive C:

When a large file file from my initiator (like 26 gig) is copied to my ISCSI img file it starts out telling me the speed is like 188 MB/s but then it starts slowing down. It appears that the copy is using only about 10% on each of the gigabit MPIO connections. The speed indictor on my copy slows down to about 20-25 MB/s write.

Other ISCSI systems I've tested start out using more of the initiators bandwidth (50% or more) and consistently use it until the complete.

With the hardware setup I can usually get > 100 MB/s writes from my Drive C: to my ISCSI drive using Open-e and starwind.

Anything I can tweak to get those write speed up?

Thanks!

Does iStorage use Block I/O or File I/O? I suspect its using Block I/O because other ISCSI targets I've tested get much better performance using File I/O.

Thanks!

Thank you for your interst, I will test the issue...

Doing some more testing, I use 33% utilization of the a single connection during a read, but writing that same file back the gigabit nic only uses 2-3% of the utilization. Definitely something going on with writes. Very puzzling.

I installed ISCSI Cake and was able to get over 50% utilization of the NIC both on reads and writes, but then again, that server allows setting cache on both server and initiator side.

I have the x64 bit trial version loaded and anxiously awaiting the results of your testing..

iStorage Server use some memory for cache, so when you copy small files in to it, the data will be written to memory, so it is very fast, but when the file is very large, the effects of the cache will be very small.
After I've tested in the same environment, It looks really strange in GB and 10GB Ethernet environment, the read operation more quickly than the write operation, I've report the issue to our R&D, we will improve the performance in the recent updates.

iStorage Server is using block I/O for some reason, we have also tested it with file I/O, it does better performance in Gigabit Ethernet but nothing changed in 100M bit Ethernet.

Maybe make Block i/o the default and File i/o an option?

Would be nice if amount of cache (at initiator and target) was settable. My test istorage server has 12 gig of cache and its sitting there not being used.

Thanks!