Hyperthreading hurts system performance

  • Thread starter netfreak
  • Start date
  • Replies: Replies 4
  • Views: Views 2,006
N

netfreak

Regulars
Messages
2,229
Location
NA
ISP
Airtel GPRS EDGE + VSNL
http://blogs.msdn.com/slavao/archive/2005/11/12/492119.aspx


Our customers observed very interesting behaviour on high-end HT-enabled hardware. They noticed that in some cases when high load is applied SQL Server CPU usage increases significantly but SQL Server performance degrades.' Another developer, this time from Citrix, was just as blunt. 'It's ironic. Intel had sold hyperthreading as something that gave performance gains to heavily threaded software. SQL Server is very thread-intensive, but it suffers. In fact, I've never seen performance improvement on server software with hyperthreading enabled. We recommend customers disable it.
 
T

thoufique

Regular
Messages
169
Location
Iasqa
ISP
Trent
.
 
Last edited:
N

netfreak

Regulars
Messages
2,229
Location
NA
ISP
Airtel GPRS EDGE + VSNL
It is related to bandwidth.

Multiple threads share same memory contrller and even the same L2 cache.

S, the only situation where HT is going to be benefetial is where you have a fewthreads staved for I/O and a smaller number of COU intensive threads. If most of the threads are CPU intensive it might have an adverse impact.

And it is a security issue as well. Two threads share the same cache and if they do not belong to the same process then this is a security nightmare.

http://news.zdnet.co.uk/hardware/chips/0,3...39198583,00.htm
 
M

max

Regulars
Messages
2,750
Location
NA
ISP
MTNL
ok....the security issue with threads sharing the same cache is based on unfounded fears. The performance problem isn't due to hyperthreading. The problem is due to bottlenecks created in the System Bus.
 
N

netfreak

Regulars
Messages
2,229
Location
NA
ISP
Airtel GPRS EDGE + VSNL
Well, disabling HT did improve the performance for Citirx and MS-SQL. So the problem is definetly related to HT.

And for security, have a look at http://www.daemonology.net/papers/htt.pdf.

This paper discusses how a rogue process can monitor OpenSSL for sensitive data.
 

Similar threads