Thanks a lot for all the replies. Will post again after the tests On Tue, May 19, 2009 at 1:56 AM, Jim Idle <[email protected]> wrote:
> > aarcee74 aar wrote: > > Thanks Jim for the reply , Yes i did had a look at the settings , but > > i assume that is more towards AIX5.2 and 5.3 has many changes as > > compared to 5.2, especially with memory handling.So i was just > > referring whether any other changes are reqd with 5.3 , related to > > jBASE . > Yeah, but a lot of the information generally holds true. > > > > > In the case of SAN, we use EMC, symmetrix and Clariion disks and i am > > sure that the performance is pretty good if we take jBASE out of the > > picture. A simple tar and compress itself is giving me 60-70MB /s and > > normal "dd" commands also giving me asimilar throughput. I need to > > verify everything from the server side , before going to EMC for > > clarifications...! > That suggests that you need to do a lot more tuning of the Clariion > system to be honest. 70MB/s is pretty weak for all that hardware. I have > tow $180 SSD drives on my Linux system and I get up to 160MB/s without > any tuning. If that's the best the SAN can do, then I would not use it. > To be honest I have not seen anything but performance and reliability > issues with SANs (regardless of jBASE). > > > > Is there any direct relationship to the modulo which is being used to > > create a file and the I/O which it makes when the file is accessed ? > Remember that your tar and dd are the fastest way you can read and write > to the file system as they produce sequential IO, which quickly turns in > to track reads. So, no database accessed in any random manner will get > anywhere close to sequential track read performance. The best you can do > is to write a small program in jBC that does a SELECT (a jBC SELECT not > an EXECUTE) then READNEXT through the file in a tight loop. The drivers > should pick that up as sequential IO. > > Now, how the SAN translates this into its own access patterns is > anyone's guess. It is a riddle wrapped in a mystery inside an enigma. As > far as I have been able to determine in the past, sequentiality of disk > blocks does seem to hold somewhat, but all the different layers in the > system significantly blur what is going on. > > In short, before even worrying about jBASE you need to download one of > the disk performance testing programs, compile it on AIX and get a > baseline for sequential and random IO. Then try the same on a local > disk/disk array. Once you have that, then you can tune the SAN in > whatever ways it allows (but lots of local RAM given to disk buffers is > going to help). Having done that, then you can test the kinds of things > your program does. Make sure that your files are well sized (which seems > to mean not using jrf on jBASE 5 :-(. > > > > Also, one more query, we have found that the jBASE/T24 is not making > > use of SMT, the second logical thread for the CPU. Is it that > > something needs to be compiled again. > > > See Pat's earlier reply about jBASE. > > > > > --~--~---------~--~----~------------~-------~--~----~ Please read the posting guidelines at: http://groups.google.com/group/jBASE/web/Posting%20Guidelines IMPORTANT: Type T24: at the start of the subject line for questions specific to Globus/T24 To post, send email to [email protected] To unsubscribe, send email to [email protected] For more options, visit this group at http://groups.google.com/group/jBASE?hl=en -~----------~----~----~----~------~----~------~--~---
