Summary Re: Solaris Virtualization

From: <>
Date: Mon Apr 12 2010 - 18:47:17 EDT
Hi All

Thanks everybody for the reply. Everybody was unanimous that "we can not standardize on single virtualization strategy". 

Another thing was not to compare T2K or below T2K with T5xxx family in other words T1/T2 chips with T2+ chip. There is a big difference in terms of speed between those.

Having said that, T or M - again you can not standardize. T-series servers are throughput based server so they are good for things like Java and Web. M-series server are good for compute intensive task. Chris Hoogendyk puts it very beautifully : -

"Think of it like a moving truck versus a sports car. If you have a single package that you need to get across the country quickly, a sports car courier might do the job quite well. If you have 1000 packages you need to get across, the sports car will take forever going back and forth. The moving truck will get them all there more quickly. The T2's are the moving trucks."

Tim Bradshaw summarizes T-series like this, which most of the people agreed with: -

"In particular, while a huge number of (virtual) processors are available, the straight-line performance of each processor is not that great. So they're great for workloads which have a lot of parallelism, but not great for things which are bounded by single-thread performance. "

Surprisingly more people actually favored use of container than LDOM, but again not to standardize.

Everybody talked about workload, so how the heck we measure it! Almost everybody pointed me to 
 Running cooltest on idle server will not help, it should be run at when server is at steady state and should be run for longer time to get accurate information.

I might be over simplifying it most of the people said Oracle, Compute job, Dataware - dont use T-series. Web, Java, Parallel computing use T-series.

My Summary: If your application is good with threads and workload test suggest that it is good candidate then put it on LDOM with T series by which I mean T2+ chip. If the cooltest returns anything other than green or you know it is single threaded app move to a container on M series server.

I can not wind up without thanking Dean Ross-Smith, Nate Gelbard, Alan, Geoffrey Walton, Brian L Jester, Rajdeep Sengupta, Chris Hoogendyk, Scott Rief, Mike Weeks, 
Victor Engle, Anton Pavlenko, Robert Milkowski, Guy Gaylord,  Jonathan Elliot, Tamer Embab, Andrew Williamson, Michael Greenberg and Przemyslaw Bak.

Thanks a lot for input. 

----- Original Message -----
Sent: 04/06/10 07:23 PM
To: Sun Managers
Subject: Solaris Virtualization

Hi Gurus

Our organization is planning to Virtualize Solaris servers to reduce the cost to the company. For that we have few proposals 

a) M-series server either use hardware partitioning or use containers.

b) T-series server for LDOM.

LDOM solution was approved because patching with containers is messy and M-series server turns out to be more expensive than T-series servers if we had to do hardware partitioning.

For T-Series our SUN rep proposed T5440. The key selling point of this hardware model is "Consolidation". And the plan is to consolidate 10 to 15 physical servers to one of these servers. So, far all sounds good. But lately I have seen many problems with cool thread servers. We migrated few of our applications from V490 and V890 to T2Ks and saw performance dropped by 5 times. And these T2Ks are not configured for LDOM. 

I worried, that if we use LDOM as our virtualization strategy, we might see some/most applications might perform badly on the virtual environment (taking a lead from T2K experience). 

1) Are my worries valid? 

2) If yes, is there a bench mark tool that I can use to decide which server should go to LDOM and which servers should be virtualized on M-series either by using hardware partition or by using containers?

sunmanagers mailing list
sunmanagers mailing list
Received on Tue Apr 13 09:56:13 2010

This archive was generated by hypermail 2.1.8 : Thu Mar 03 2016 - 06:44:16 EST