Thread Links | Date Links | ||||
---|---|---|---|---|---|
Thread Prev | Thread Next | Thread Index | Date Prev | Date Next | Date Index |
Ali
I have n’t found any references to link lengths, though they may be buried in the paper. Fortunately the SG does have contributions from Paul Kolesar providing data on the structured cabling for:
access <-> distribution layer
and
distribution <-> core
It is certainly possible, as you suggest, and the paper alludes to this, that the core may be constructed from a mesh of stackable switches rather than large modular switches. This would have the impact of increasing the number of optical interfaces used to connect the switches forming the core. Essentially such a design would trade off backplane traces for optical interfaces. It is possible of course that these interfaces would be made by patch cables (switches are in the same rack) rather than using structured cabling.
The fan out of a distribution switch in the warehouse scale computer (10 racks) is similar to the data provided by Scott Kipp in his submission the BWA adhoc:
http://www.ieee802.org/3/ad_hoc/bwa/public/may11/kipp_01_0511.pdf
Where 16 racks are connected to the distribution layer switch, called a cluster switch in this presentation.
The fact that a cluster can be housed in a 40 ft shipping container indicates how short these links can be.
Regards
Andy
From: "Ali Ghiasi" <aghiasi@xxxxxxxxxxxx>
To: STDS-802-3-100GNGOPTX@xxxxxxxxxxxxxxxxx
Sent: Monday, November 28, 2011 11:54:45 AM
Subject: Re: [802.3_100GNGOPTX] Warehouse Scale Computing, impact on reach objective
Dan
Ali,I found the following interesting. Pg 40"Datacenter sizes vary widely. Two thirds of US servers are housed in datacenters smaller than5,000 sq ft (450 sqm) and with less than 1 MW of critical power [26] (p. 27). Most large datacentersare built to host servers from multiple companies (often called co-location datacenters, or “colos”)and can support a critical load of 10–20 MW. Very few datacenters today exceed 30 MW of criticalcapacity."5000 sq ft translates to a mere 70ft by 70ft.I have to wonder where this leads us? We hear of massive data centers taking up city blocks, and then we see something like this telling us that a super-majority of data centers can fit in a medium sized McMansion.I tend to believe both, and would like to see the distribution of actual data center reach requirements. Now digging into some of the reference material :i.e. Data Center Report to CongressAndy,Thanks. This is great stuff!Regards,DanFrom: Ali Ghiasi <aghiasi@xxxxxxxxxxxx>
Reply-To: Ali Ghiasi <aghiasi@xxxxxxxxxxxx>
Date: Mon, 28 Nov 2011 10:00:08 -0800
To: 100G Group <STDS-802-3-100GNGOPTX@xxxxxxxxxxxxxxxxx>
Subject: Re: [802.3_100GNGOPTX] Warehouse Scale Computing, impact on reach objectiveAndyThank you for forwarding this report, I glanced through the report but I was not able to get any specific cable lengthdistribution from this report. Figure 1.2 is a good example of TOR architecture. Can we assume that the fromTOR to each CPU blade Cu will be used and MMF fiber will be used from TOR switch to the data center switch?The next question is what is the typical reach between TOR switch and data center switch?You also elude to more flat architecture I assume something like Clos where data center switch is replaced with distributed fabric.Do you have inside what typical cable reach will be in this case?Thanks,AliOn Nov 26, 2011, at 12:53 PM, Andy Moorwood wrote:
Study Group Members,
I share the regret, expressed in several posts to this list, that large internet data center operators are unwilling to make their requirements known in an open non confidential manner. I would like to forward to the group a paper, recommended by a colleague, that may help close this information gap.
“The Datacenter as a Computer: An Introduction to the Design of Warehouse-Scale MachinesLuiz André Barroso and Urs Hölzle”
ISBN: 9781598295566 paperbackISBN: 9781598295573 ebook
The paper may be viewed, without charge, at the Morgan & Claypool site, but please be aware of the restrictive notices on page (iv)
http://www.morganclaypool.com/doi/pdf/10.2200/S00193ED1V01Y200905CAC006
Copies may also be purchased at internet book sites.
The paper, 120 pages in all, describes the specific challenges faced when applications implemented by internet content providers, such as Google and Microsoft, require many thousands, even tens of thousands of servers. Indeed the usage model of the data center changes from “a place to house servers” to a “building to host an application”.
The introduction, pages 1 to 11, gives insight into why these warehouse computers differ from traditional data centers and how this impacts the need for communication bandwidth within the data center.
The ideal system, as described by Barosso and Hölzle, would be one where the cross sectional communication bandwidth of the data center would equal to the bandwidth of the servers, i.e. a network without over subscription. In such a system the application developer can freely locate functions throughout the network, optimally distributing load and minimizing computational and HVAC hotspots. The authors admit that economic considerations cannot support such a model and that over subscription levels of 5:1 are evident between racks of servers (80 servers per rack), 10 racks in a group (800 servers). Using the terminology as per kolesar_02_0911_NG100GOPTX, page 4, citing barbieri_01_0107.pdf, this over subscription would refer to the “access" to “distribution” network layers.
http://www.ieee802.org/3/100GNGOPTX/public/sept11/kolesar_02_0911_NG100GOPTX.pdf
Decreasing the relative cost of these access to distribution layer links would enable warehouse scale computer builders to reduce the level over subscription and get closer to their ideal system. Throughout the paper the authors use a system wide approach to find the lowest cost. By this I mean reducing cost in one area is not beneficial if it results in an increase in overall costs, since cost is just shifted from one area to the other.
Such considerations should play a part in the determination of a reach objective. As we increase the reach to include an ever higher percentage of the links as described in kolesar_02_0911_NG100GOPTX, we should be cognizant of the increase in relative cost to achieve this increased reach; and evaluate if, when considered at a network level with a distribution of link lengths as per Paul’s presentation, we are decreasing overall cost, or not.
Best Regards
Andy