A fellow member of the Oaktable Network sent out an interesting URL. So how badly do you think a Pentium 4 overclocked to 8GHz would stall on memory loads? Can you say, “CPI?”
DISCLAIMER
I work for Amazon Web Services. The opinions I share in this blog are my own. I'm *not* communicating as a spokesperson for Amazon.
In other words, I work at Amazon, but this is my own opinion.
Pages
- About
- Index of Posts
- Misc
- Papers, Webcasts, etc
- SLOB Resources
- Little Things Doth Crabby Make
Blogroll
- AWS Database Blog
- Bart Sjerps
- Beat Ramseier
- Bertrand Drouvot's blog
- Daniel Westermann
- flashdba
- Franck Pachot
- Frits Hoogland
- Gokhan Atil
- James Morle
- Jan Karremans Blog (Johnnyq72)
- Jason Arneil
- Jonathan Lewis
- Luca Canali's Blog
- Mahmoud Hatem Blog
- Mark Smith
- Martin Bach
- Mauro Pagano
- Nikolay Savvinov
- Noons
- Pardy DBA
- Real World Technologies
- StorageIOblog (Greg Shulz)
- THE ORACLE ALCHEMIST
- The SQL Herald
- Tim Hall
Join 743 other subscribers
Recent Posts
- VMware Are Serious About Oracle Database Performance Characterization
- Little Things Doth Crabby Make – Part XXIII. IEC 60027 Still Makes Me Crabby!
- Microsoft Shares Oracle Database Direct NFS Capabilities with SLOB Testing Results
- Announcing SLOB 2.5.4
- Announcing SLOB 2.5.3
- (no title)
- Following Various Legal Action Regarding “Oracle Cloud Revenue”
- Announcing SLOB 2.5.2.4
- Announcing SLOB 2.5.2.3
- Announcing SLOB 2.5.2.2
Recent Comments
11.2.0.4
Automatic Workload Repository
AWR
columnar database
column projection
DBaaS
Exadata
Exadata benchmarks
Exadata ERP
Exadata OLTP
Exadata Storage Grid
Exadata Xeon 5600 Datawarehousing
HP Oracle Database Machine
Netezza
Netezza Oracle
NUMA
numactl
OOW12
OpenWorld
OpenWorld 2011
Oracle
Oracle Database performance XtremIO flash
Oracle Exadata Storage Server
Oracle Exadata Storage Server Software
Oracle I/O Performance
Oracle Performance
Oracle Programmable Storage Server
predicate offload processing
Random I/O
Sandy Bridge
SLOB
SLOB Testing
SUMA
Teradata
whitepaper
X2-2 vs X2-8
Xeon E5
Xeon E5-2600
Xeon E7 Performance
XtremIO
Copyright
All content is © Kevin Closson and "Kevin Closson's Blog: Platforms, Databases, and Storage", 2006-2015. Unauthorized use and/or duplication of this material without express and written permission from this blog’s author and/or owner is strictly prohibited. Excerpts and links may be used, provided that full and clear credit is given to Kevin Closson and Kevin Closson's Blog: Platforms, Databases, and Storage with appropriate and specific direction to the original content.
quick request for a blog entry:
I know you address it somewhat in the book, but here goes anyway:
in a NUMA multi-quad, don’t you inherently get a high CPI whenever there is a mem sync clash across quads? How does that impact on things like latches, for example? I never understood properly that section of the book and since you might know a thing or two about the subject I thought I’d ask? 😉
Hi Noons,
Yes, I’ll need to discuss coherency misses. It all depends on how latches are implemented. In short, using a simple test and set sort of spinlock, NUMA suffers with heavily contended latches. In the end, however, it all depends on how “lumpy” the memory is. If a remote memory write is some 100x more costly than a local memory write, then normal spinlock just don’t work–well, they don’t fail but they don’t scale 🙂
I’ll make sure to cover some of those goodies.