LRDIMMs, RDIMMs, and Supermicro's Latest Twin
by Johan De Gelas on August 3, 2012 4:45 AM EST- Posted in
- IT Computing
- Intel
- Samsung
- Xeon
- Cloud Computing
- Supermicro
The Purpose of this Test
We will not state that we had the most optimal testing configuration for our objectives. The most interesting LRDIMMs are the 32GB ones, and we had the cheaper (and thus easier to borrow) 16GB parts. The most interesting Supermicro Server is the 6027TR-D70RF+; we had the slightly older but very similar 6027TR-D71FRF. As always, we try to make the best with what we have in the lab. We believe that with this testing configuration we can still answer the questions that will pop up when you consider the different server configuration such as:
- How much bandwidth and latency will you sacrifice when buying LRDIMMs instead of RDIMMs?
- Does investing in expensive high capacity DIMMs pay off?
- Can you really host twice as many VMs in twice as much memory?
- How much performance do you gain by giving your VMs more physical memory?
Those two last questions seem silly at first sight but they are not. In most cases, virtual machines get (much) more memory than they really need. Most administrators prefer to give the VM quite a bit of memory headroom. It is a good practice, but it means that it is not necessarily a bad thing if the total amount of virtual memory is quite a bit higher than the total available physical memory. Unless all VMs are working hard, a modern hypervisor can make sure that the "needy" VMs get what they need and the "lazy" ones get only the bare minimum of physical RAM.
Indeed, an advanced hypervisor such as ESXi (especially ESXi) has a lot of tricks up its sleeve to make sure that even if you don't have enough physical memory, your VM will still run fine. Physical memory use is optimized by:
- Transparent page sharing (TPS): the Hypervisor will only claim one page for several pages of several different VMs with identical content (e.g. the Windows kernel and HAL in several Windows based VMs)
- Ballooning: the Hypervisor reclaims memory that a VM does not use and gives it to more "needy" VMs
- Memory compression: pages that need to be swapped out (see further) are checked if they have a high compression ratio. If that is the case they are compressed and kept in a memory cache. As a result...
- Hypervisor swapping: memory that is not active and not compressible can be swapped to disk, similar to a "normal" OS (Supervisors). This does not necessarily result in a large performance hit, as pages swapped out are rarely used.
So we thought it would be interesting to design a scenario where we could measure the performance differences between a system with lots of memory and a more budget limited one.
Facebook
Twitter
RSS
26 Comments
View All Comments
dgingeri - Friday, August 3, 2012 - link
"Most 2U servers are limited to 24 memory slots and as a result 384GB of RAM. With two nodes in a 2U server and 16 slots per node, you get cram up to 512GB of RDIMMs in one server. "It's not one server. It's actually 2 servers. just because they're in a 2U X 1/2 width form factor doesn't mean they're just one system. There are 2 systems there. Sure you can pack 512GB into 2U with 2 servers, but there are better ways.
1. Dell makes a PowerEdge R620, where you can pack 384GB into 1U, two of those gives you the same number of systems in the same space, with 50% more memory.
2. Dell also has their new R720, which is 2U and has a capacity of 768GB in a 2U form factor. Again, 50% more memory capacity in the same 2U. However, that's short 2 processor sockets.
2. Now, there's the new R820. 4 sockets, 1.5TB of memory, 7 slots, in 2U of space. It's a beast. I have one of these on the way from Dell for my test lab.
Working as an admin in a test lab, dealing with all brands of servers, my experiences with various brands gives me a rather unique insight. I have had very few problems with Dell server, despite having nearly 30% Dell servers. We've had 7 drives die (all Toshiba) and one faceplate LCD go out. Our HP boxes, at less than 10% of our lab, have had more failures. The IBMs, ahile also less than 10%, have had absolutely no hardware failures. Our Supermicros comprise about 25% of the lab, yet contribute >80% of the hardware problems, from motherboards that just quit recognizing memory to backplanes that quit recognizing drives. I'm not too happy with them.
JHBoricua - Monday, August 6, 2012 - link
Dgingeri,Sure, you can load each of those Rxxx Dell servers with boatloads of memory, but you fail to mention that it comes with a significant performance/penalty. The moment you put a third Dimm on a memory channel your memory speeds drops from 1600 (IF you started with 1600 memory to begin with) to 1066 or worse, 800. On a virtualization host, that makes a big difference.
Casper42 - Friday, August 10, 2012 - link
No one makes 32GB @ 1600 yet.So 512GB @ 2DPC would be 1333
And 768GB @ 3DPC would be 1066 or 800 like you mentioned.
384 using 16GB DIMMs would still be 3DPC and would drop from 1600 down to like 1066.
256GB @ 1600 @ 2DPC still seems to be the sweet spot.
BTW, why is the Dell R620 limited to 16GB DIMMs? The HP DL360p Gen8 is also 1U and supports 32GB LRDIMMs
ImSteevin - Friday, August 3, 2012 - link
MMhmmm yeahOh yeah ok
I know some of these words.
thenew3 - Friday, August 3, 2012 - link
The latest Dell R620's are 1U servers that can have two 8 core CPU's and 24 DIMM slots. Each slot can hold up to a 32GB DIMM giving total memory capacity of 768GB in a 1U space.We use these in our data centers for virtualization (we're 100% virtualized). Completely diskless (internal RAID 1 dual SD modules for ESXi)
Each machine has four 10gb NIC plus two 1gb NIC. All storage on iSCSI SAN's through 10gb backbone.
For most virtualization tasks, you really don't need the 2U R720, which has the same CPU/RAM options but gives you more drive bays and expansion slots.
shuntian8099 - Saturday, August 4, 2012 - link
We accept any form of payment.
http://commonprosperity.org/ )
S=h=0=x ==sh0es=4oUSD
J=0=r=d=a=n==sh0es=36USD
A=i=r==M=a=x=sh0es=39USD
C=0=a=c=h==b=a=g=36USD
S=u=n=g=l=a=s=s=15USD
simpletrading@hotmail.com
Finally (48 hours) time limit to buy.
LV Muffler $ 5.99
LV Bags $ 19.9
LV Wallet $ 6.55
Armani Glasses $ 5.99
LV Belt $ 6.9
simpletrading@hotmail.com
Buy addresses-
commonprosperity.org
╭⌒╮
∴★∵**☆．∴★∵**☆．∴★∵**☆．█████.::∴★∵**☆．∴★∵**☆．█田田█::
commonprosperity.org ╭⌒╮ ╭⌒╮
∴★∵**☆．∴★∵**☆．█田田█.∴★∵**☆．∴★∵**☆．█田█田█∴★∵**☆．∴★∵*
*☆．█田█田█.★∵**☆．∴★∵**☆．█████．*******************You
r satisfaction is our eternal pursuit!
ddr3memory - Sunday, August 5, 2012 - link
A few corrections - the 192GB for HCDIMMs is incorrect - it should also be 384GB.There is no data available that confirms a 20% higher power consumption for HCDIMMs over LRDIMMs. There is a suspicious lack of benchmarks available for LRDIMMs. It is possible that figure arises from a comparison of 1.5V HCDIMMs vs. 1.35V LRDIMMs (as were available at IBM/HP).
It is incorrect that LRDIMMs are somehow standard and HCDIMMs are non-standard.
In fact HCDIMMs are 100% compatible with DDR3 RDIMM JEDEC standard.
It is the LRDIMMs which are a new standard and are NOT compatible with DDR3 RDIMMs - you cannot use them together.
The 1600MHz HCDIMM mention is interesting - would be good to hear more on that.
ddr3memory - Sunday, August 5, 2012 - link
I have posted an article on the performance comparison of HyperCloud HCDIMMs (RDIMM-compatible) vs. LRDIMMs (RDIMM non-compatible).Cannot post link here it seems - search for the article on the ddr3memory.wordpress.com blog:
Awaiting 32GB HCDIMMs
ddr3memory - Monday, August 6, 2012 - link
VMware has had good things to say about HCDIMM (not a word from VMware about LRDIMMs though). Search on the net for the article entitled:Memory for VMware virtualization servers
ddr3memory - Monday, August 6, 2012 - link
The prices mentioned maybe off - I see IBM showing same retail prices for 16GB LRDIMMs/HCDIMMs and similar at the IBM resellers.These resellers show 16GB HCDIMMs selling at $431 at costcentral for example, $503 at glcomp and $424 at pcsuperstore.
Search the internet for this article:
What are IBM HCDIMMs and HP HDIMMs ?
It has the links for the IBM/HP retail prices as well as the reseller prices.