andyvasey Posted June 21, 2007 Posted June 21, 2007 Hi, i am new to this orumn and have an issue with both xp 64 bit an server 2003 64bit. Please have a read through the following and drop me a line if you have and ideas?Many ThanksAndy andy@five-9s.net I am currently working on a large project but have hit a huge stumbling block and believe it to be down to a limitation or setting within the 64Bit Microsoft O/S. The basic issue arises when trying to packets of data being read from the disk are 32MB or higher, if 31.99MB or less then the throughput remains constant but at 32.01MB the data transfer literally stops dead. I have tested this now on several drives and controllers and have found this to be the same across the board unless using a 32BIT O/S and the cut off here is 64MB. The following is some further specific information on the testing.**Please see the test report below. The basic issue is that if the file transfer frames are 32MB or above the throughput drops from 1GB to 252MB ?? I think this limitation may be down to the disks?******************************** A 4K image file on disk is ~35 MB. There are larger "4K" formats (different pixel widths/height or bit depths etc.) : 4096 x 2160 pixels x 4 bytes == ~35 MB files 4K RAM looping --- 29 fps (steady)We need a minimum of 24 fps, so this is a good result.But : 4K off disk to screen -- 2 fpsThis was such a terrible result that we had to look deeper.I know 2K is 60 fps.Using iometer (see below), I noticed a significant issue as we crossed the 32 MB frame size boundary ie. frame size 12 MB --- 1 GB/sec off disk frame size 16 MB --- 1 GB/sec off disk frame size 24 MB --- 1 GB/sec off disk frame size 31 MB --- 1 GB/sec off disk frame size 32 MB --- ** I/O errors off disk frame size 35 MB --- ** I/O errors off diskOur application currently reads the whole image file off disk to display on screen, so this request for >32 MB would appear to cause the performance hit.Interestingly - 4096x1800 images (29 MB each) give 26 fps off disk to screen.We are now looking at a pure RAID-0 configuration on 2 volumes (as a baseline).Adding the ATTO configuration setting RGReadSpeed on each volume (a read ahead), added extra performance - but there us still a problem.4096x2160 (35 MB) -- 1x RAID-0 -- ------------ 8/9 fps 4096x2160 (35 MB) -- 2x RAID-0 -- XP striped - 8/9 fps 4096x1800 (29 MB) -- 2x RAID-0 -- XP striped - 29 fps (*)(*) this might be 26 rather than 29.So, this is not just iometer behaving strangely, our "real world" program sees some issue at this point.**If I create "JBOD" containing a single disk - 1 primary partition and NTFS filesystem, I get the same behaviour with iometer as before.1 SAS disks JBOD (700 GB) NTFSiometer frame sizes vs read speed1) 1 MB 75-80 MB/s2) 1 MB 900 KB 75-80 MB/s3) 2 MB 64-77 MB/s4) 12 MB 600 KB 40-90 MB/s5) 31 MB 30-90 MB/s6) 32 MB ERRORThe I/O gets much more unstable as we increase the request size.We get the break at the same place with a single disk. So this does not appear to be stripe related.iometer gives the same error : Error 1450 reading 33554432 bytes from disk S:\ Error performing I/O to S:\www.five-9s.net
cluberti Posted June 26, 2007 Posted June 26, 2007 Please don't double-post. It's against the forum rules:» Posting Guidelines2.a Use some common sense. Don't post your question under multiple topics. Think before posting and choose the right topic for you question or answer. When posting, have something to say, not just "My First post!" or "Cool site!!!" We sort of have the idea by now :-).
Recommended Posts
Create an account or sign in to comment
You need to be a member in order to leave a comment
Create an account
Sign up for a new account in our community. It's easy!
Register a new accountSign in
Already have an account? Sign in here.
Sign In Now