[rhn-users] Intel or AMD

Dean WOODS dean.woods at st.com
Thu Dec 16 22:52:16 UTC 2004


For what it's worth, we use the Sun V20Z Opteron systems and the NewIsys 
2100 Opteron systems.  Sun actually OEM's the NewIsys 2100 as the V20Z.  
If your on a tight budget, the NewIsys system is quite a bit cheaper 
than the V20Z for the same options.  Both are great machines and perform 
very well.  The only real difference I see between them is the way they 
utilize the service ports for remote access and management.  For more 
info and pricing, check out NewIsys 
<http://www.newisys.com/products/2100.html> and Colfax International. 
<http://colfax-intl.com/jlrid/SpotLight.asp?IT=0&RID=71>

DWoods

Dominique Demore wrote:

>Red Hat Network Users List <rhn-users at redhat.com> on Thursday, December 16, 2004 at 13:28 -0500 wrote:
>  
>
>>Single CPU?  Dual CPU?  Quad CPU?
>>    
>>
>
>Well I am hoping to be purchase a Dual 248 or 250 version of the Opteron. However since Oracle's licensing is pretty
>hefty for the second processor, I will have to wait an see.
>
>As for the applications, these servers would only be for the SIS database and Application Server (plus any custom apps
>which access Oracle). You mentioned the HP Server. I was looking at their prolient DL models.
>
>  
>
>>Opteron's use a Non-Uniform Memory Architecture (NUMA).  Specifically, 
>>each CPU has its own connection to its own bank of RAM.  Memory is 
>>accessed via a different path than I/O.  Then there's connections 
>>between the CPUs in case one CPU needs to access RAM or I/O hooked up to 
>>the other processor.  Xeons and Itaniums have a Front Side Bus that all 
>>access to all I/O and memory goes through, leaving you with a single 
>>bottleneck.
>>
>>
>>So, it depends on the exact Opteron motherboard design that you're 
>>talking about, but with a dual or quad CPU system, an Opteron will make 
>>a much better database server than a Xeon or an Itanium (assuming the 
>>motherboard is designed right.)  I've looked at the design specs for 
>>Sun's Opteron servers and HP's Opteron servers, and both are solid 
>>designs that put RAM in the right places.  In both cases, on the 
>>dual-opteron design all I/O is off of one CPU (with ram on both) and on 
>>the system that can handle more than 2 CPUs, I/O is spread between 2 of 
>>the CPUs.  (and a bank of RAM on every CPU, of course)  I haven't 
>>checked if they've updated, but about a year ago IBM's Opteron systems 
>>looked kind of sub-par (no on-board RAID); designed for compute clusters 
>>not database servers.  I have no idea about other vendors; quality can 
>>vary widely.  There are definitely some Opteron boards out there that 
>>put all the RAM and I/O on one CPU, which makes the system about as slow 
>> as if it had a front-side bus.
>>
>>
>>Check for support from Oracle.  An Opteron can still make a very speedy 
>>server for running 32-bit applications, but you're best off if the 
>>specific things you're running actually support the "x86_64/AMD64/ia32e" 
>>64-bit instructions.
>>
>>
>>Also, go read this article, and make sure to look at the diagrams:
>>http://www.samag.com/documents/s=9408/sam0411b/0411b.htm
>>-- 
>>Eric Eisenhart <eric.eisenhart at sonoma.edu>
>>Linux/Unix Systems Administrator
>>Office: Schulz 1050A, (707) 664-3099
>>AIM: ericeisenhart, ICQ: 156218985
>>Sonoma State University, IT
>>
>>
>>_______________________________________________
>>rhn-users mailing list
>>rhn-users at redhat.com
>>https://www.redhat.com/mailman/listinfo/rhn-users
>>    
>>
>
>
>
>-----
>Dominique Démoré
>Technical Services Coordinator
>Rainbow District School Board
>69 Young Street
>Sudbury, Ontario
>P3E 3G5
>Tel: (705) 674-3171 x. 258
>Fax: (705) 671-2442
>
>
>_______________________________________________
>rhn-users mailing list
>rhn-users at redhat.com
>https://www.redhat.com/mailman/listinfo/rhn-users
>  
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listman.redhat.com/archives/rhn-users/attachments/20041216/f4107a03/attachment.htm>


More information about the rhn-users mailing list