Rick Todd (4/19/2010)

---

I think there's a significant element missing from the "which way is faster to process" debate (hash then compare, or just compare): time to market. As in, how...

Nadrek (4/19/2010)

---

Mike C (4/17/2010)

---

Not sure how you calculated the probability of collision at 2^40 off the top of your head... You can engineer an MD5 collision in 2^39 or...

tarquin (4/19/2010)

---

Here I was believing in Kimball! To prove or disprove my guru's theory I went and looked at a client database that has been in operation for about a...

Tom.Thomson (4/17/2010)

---

Mike C (4/17/2010)

---

The vast majority of ETL solutions I've encountered tend to discard a lot more duplicate rows than they update or insert (after the initial table load, that...

Here's a...

LOL. ...

Tom.Thomson (4/17/2010)

---

Mike C (4/17/2010)

---

This is why I recommend SHA-1 or better (probability of collision for 160-bit hash is 2^80 per the birthday paradox). Even the current crop of theoretical...

Tom.Thomson (4/17/2010)

---

Mike C (4/17/2010)

---

This is why I recommend SHA-1 or better (probability of collision for 160-bit hash is 2^80 per the birthday paradox). Even the current crop of theoretical...

Tom.Thomson (4/17/2010)

---

Mike C (4/14/2010)

---

I wouldn't expect many collisions with MD5 unless you're processing 2^64 or more rows. If you're getting collisions with MD5 or SHA I would immediately suspect...

Tom.Thomson (4/17/2010)

---

Mike C (4/14/2010)

---

tarquin (4/14/2010)

---

Wrapping hashbytes with checksum is not bad as long as you include all columns in your join as indicated and create the index including all columns...

tarquin (4/16/2010)

---

Quoting "The Microsoft Data Warehouse Toolkit" by Ross and Kimball pg 245: "...This is a common technique for high performance dimension change management. Compute the hashes and store them...

tarquin (4/15/2010)

---

I dont agree with this. An index is only useful for...

You'll have to check with the folks over at njsql (njsql.org). They're the keepers of the schedule.

Sean Terry (4/14/2010)

---

Mike C (4/14/2010)I also prepend the binary length of the string to each column, which provides additional differentiation and provides a useful method for dealing with NULLs. ...

As long as you choose cryptographically secure hashes with long bit lengths (160 bits or more recommended) you'll do great. Where to hash depends on where you can do...