Comments for SQL Server Input/Output

Comment on SQLDIY: Gather Virtual File Statistics Using T-SQL #TSQL2sDay 15 by Jason

Jason — Mon, 23 Apr 2012 13:57:19 +0000

Wes, you are my hero. You are definitely the leader in the industry. It is wonderful to see you and attend your sessions at SQL Saturady #107 Houston.

Comment on Speaking In April #SQLSat111 and #SQLSat107 by Patrick

Patrick — Mon, 16 Apr 2012 16:52:41 +0000

Hi, Wes, I thought your presentation was fantastic, and I got a lot out of it. Do you plan on making the slide deck available?

Comment on SQLDIY: Gather Virtual File Statistics Using T-SQL #TSQL2sDay 15 by Who Has Busy Files? Identifying Performance Hogs

Who Has Busy Files? Identifying Performance Hogs — Mon, 19 Mar 2012 14:58:03 +0000

[...] There are many versions out on the web, but I found Wes has a particularly good version over here, which I [...]

Comment on SQLDIY: Gathering A Data Dictionary by Bob Atkinson

Bob Atkinson — Mon, 19 Mar 2012 03:56:07 +0000

Nice to see this, and especially from you.
NASA had big buy-in for similar techniques, and received good value as questions arose and answers were handy. Great value for new personnel on existing project. Also strong contribution for online-help, product documents, etc.

Comment on My Tools by Stefan

Stefan — Wed, 22 Feb 2012 09:24:40 +0000

Hi Wes,
Excellent work on UBCopy!….and It is working on itanium as well :-)

One question though, is the latest release really 1.5.2.1851? when looking at https://github.com/SQLServerIO/UBCopy the latest version mentioned seems to be 1.7.41.24. However the file that is available for download is 1.5.2.1851.

If possible I really would appreciate if I can download release 1.7.41.24.

Regards,
Stefan

Comment on At The End of the IO Road With C#? Pave New Road! by Stefan

Stefan — Wed, 22 Feb 2012 09:13:06 +0000

Hi Wes,

Excellent work!….and It is working on itanium as well :-)

If possible I really would appreciate if I can download release 1.7.41.24.

Regards,
Stefan

Comment on News Flash! PASS Board Makes Unpopular Decision! by Emalclumsoica

Emalclumsoica — Wed, 15 Feb 2012 03:44:12 +0000

Bravo, questa grande idea appena inciso

Comment on Fundamentals of Storage Systems – Stripe Size, Block Size, and IO Patterns by Wes Brown

Wes Brown — Sun, 22 Jan 2012 14:27:34 +0000

raid 10 always assumes nX for reads and n/2 writes. Some vendors will read from both disk some will choose lowest latency return so you need to test that.

Again,
the cache hit for reads/writes is causing your problems. If you set cache hits to 0 you get the expected results.

If you want to see the equations just right click and view source on the page the are all there.

Comment on Fundamentals of Storage Systems – Stripe Size, Block Size, and IO Patterns by Yuval Cohen Hadad

Yuval Cohen Hadad — Sun, 22 Jan 2012 06:37:29 +0000

Hi Wes,

thansk for the prompt reply.
i changed the cache numbers to 50% and got your numbers
and then return back to zeros to be close to real world numbers.

I am in the middle of designing a big SAN space, 45*300GB 15K SAS, for SQL server 2008 datawarehouse and your calculator is very very helpul! thanks you!

but i am struggling to understand the RAID results and what impact them.

just two exmple:
1. in Read Seq. pattern, the numbers of RAID 0 is always double the numbers of RAID 10. shouldn’t RAID 0 and 10 has the same read performance (nX)?
2. Seq. read IO should be faster than Random Read IO and in the “Disk Results” its actualy so. but when i changed the IO pattern to “both” i get much better results. isnt “Both” a combination of Random and Seq.? and if so should be less faster than a Seq. only case?

Can you share Peter Chen ‘ equation for RAID 10 ?

Thanks a lot, Yuval

Comment on Fundamentals of Storage Systems – Stripe Size, Block Size, and IO Patterns by Wes Brown

Wes Brown — Sat, 21 Jan 2012 17:47:54 +0000

Hi,
Thanks for the comments. The article is poorly worded on what I used to plug into the raid calculator. if you put a read cache of 50% and a write cache of 50% you get the numbers I state in the article. If you put it at 0/0 it gives you the numbers that are much closer to the real world testing examples I show in the charts.

I’ll make some revisions to make it easier to follow.