Re: AWS Enhanced Networking

From: Rayson Ho <no email>
Date: Thu, 9 Jan 2014 20:25:26 -0500

As far as I know, not yet. I just created my own AMI using
non-standard procedure...

Both the Enhanced Networking feature & the new GPU instance family
require new drivers / AMIs. Thus if one wants to take full advantage
of the new features in the c3, i2, and g2 instance families, a new set
of AMIs (with the Intel or NVidia drivers installed) need to be
created.

I just came back from my vacation a few days ago, and I will need to
get back on the SC IRC and see what remaining work needs to be done.

Rayson

==================================================
Open Grid Scheduler - The Official Open Source Grid Engine
http://gridscheduler.sourceforge.net/
http://gridscheduler.sourceforge.net/GridEngine/GridEngineCloud.html


On Thu, Jan 9, 2014 at 7:59 PM, Sergio Mafra <sergiohmafra_at_gmail.com> wrote:
> Hi Rayson,
>
> Is there any AMI with the Intel ixgbevf Virtual Function drive ready for
> StarCluster?
> I think that this should be part of StarClusterīs public AMIs.
>
> All best,
>
> Sergio
>
>
> On Wed, Jan 8, 2014 at 3:52 PM, Rayson Ho <raysonlogin_at_gmail.com> wrote:
>>
>> The Enhanced Networking feature provides much better latency for all
>> instance types that support it. For the c3.8xlarge instance type,
>> Enhanced Networking allows the instance to utilize over 95% of the
>> 10Gbit Ethernet bandwidth:
>>
>>
>> http://blogs.scalablelogic.com/2013/12/enhanced-networking-in-aws-cloud.html
>>
>> http://blogs.scalablelogic.com/2014/01/enhanced-networking-in-aws-cloud-part-2.html
>>
>> Rayson
>>
>> ==================================================
>> Open Grid Scheduler - The Official Open Source Grid Engine
>> http://gridscheduler.sourceforge.net/
>> http://gridscheduler.sourceforge.net/GridEngine/GridEngineCloud.html
>>
>>
>>
>> On Mon, Nov 25, 2013 at 12:42 PM, Rayson Ho <raysonlogin_at_gmail.com> wrote:
>> > Thanks Sergio, I check the docs the day when c3 was announced, there's
>> > no mention of placement group support for c3. I think I was reading
>> > the older version of the doc.
>> >
>> > Yuichi, we can't take advantage of SR-IOV (yet) as we don't run in a
>> > VPC.
>> >
>> > Rayson
>> >
>> > ==================================================
>> > Open Grid Scheduler - The Official Open Source Grid Engine
>> > http://gridscheduler.sourceforge.net/
>> > http://gridscheduler.sourceforge.net/GridEngine/GridEngineCloud.html
>> >
>> >
>> > On Mon, Nov 25, 2013 at 9:39 AM, Yoshiara, Yuichi
>> > <yoshiara_at_amazon.co.jp> wrote:
>> >>
>> >> Hi Sergio,
>> >>
>> >>
>> >>
>> >> There is a new feature called Enhanced Networking which is available in
>> >> C3 instances.
>> >>
>> >> Please read through the link below and see if it helps you.
>> >>
>> >>
>> >>
>> >>
>> >> http://docs.aws.amazon.com/AWSEC2/latest/UserGuide/enhanced-networking.html
>> >>
>> >>
>> >>
>> >>
>> >> http://aws.amazon.com/ec2/faqs/#What_networking_capabilities_are_included_in_this_feature
>> >>
>> >>
>> >>
>> >> Yuichi Yoshiara
>> >>
>> >>
>> >>
>> >>
>> >>
>> >> From: starcluster-bounces_at_mit.edu [mailto:starcluster-bounces_at_mit.edu]
>> >> On Behalf Of Sergio Mafra
>> >> Sent: Monday, November 25, 2013 8:16 PM
>> >> To: Rayson Ho; starcluster_at_mit.edu
>> >> Subject: Re: [StarCluster] MIT StarCluster and the new C3 instances
>> >>
>> >>
>> >>
>> >> Hi Rayson,
>> >>
>> >>
>> >>
>> >> I noticed that StarCluster didnīt allocate a placement group for C3
>> >> instances and did it by myself in the code.
>> >>
>> >> But this was not helpfull since C3 instances uses high throughput
>> >> insted of the 10 giga network.
>> >>
>> >> In my tests, using C3 or CC2 in a cluster config where MPI is required
>> >> will be equal to same processing time.
>> >>
>> >> Hope that AWS put them on 10 Giga net in the near future.
>> >>
>> >>
>> >>
>> >>
>> >>
>> >> Compute optimized
>> >>
>> >> c3.8xlarge
>> >>
>> >> 64-bit
>> >>
>> >> Intel Xeon E5-2680
>> >>
>> >> 32
>> >>
>> >> 108
>> >>
>> >> 60.00
>> >>
>> >> 640 GB (2 x 320 SSD)
>> >>
>> >> -
>> >>
>> >> 240
>> >>
>> >> High
>> >>
>> >> Compute optimized
>> >>
>> >> cc2.8xlarge
>> >>
>> >> 64-bit
>> >>
>> >> Intel Xeon E5-2670
>> >>
>> >> 32
>> >>
>> >> 88
>> >>
>> >> 60.50
>> >>
>> >> 3370 GB (4 x 840)
>> >>
>> >> -
>> >>
>> >> 240
>> >>
>> >> 10 Gigabit
>> >>
>> >>
>> >>
>> >>
>> >>
>> >>
>> >>
>> >>
>> >>
>> >> On Mon, Nov 18, 2013 at 7:13 PM, Sergio Mafra <sergiohmafra_at_gmail.com>
>> >> wrote:
>> >>
>> >> Hi Rayson,
>> >>
>> >>
>> >>
>> >> Iīll try to do that.
>> >>
>> >>
>> >>
>> >> All best,
>> >>
>> >>
>> >>
>> >> Sergio
>> >>
>> >>
>> >>
>> >> On Mon, Nov 18, 2013 at 1:27 PM, Rayson Ho <raysonlogin_at_gmail.com>
>> >> wrote:
>> >>
>> >> It's in the development branch:
>> >>
>> >> https://github.com/jtriley/StarCluster/pull/325
>> >>
>> >> It's a few lines of code, and you can merge the changes into your
>> >> local version very easily.
>> >>
>> >> Rayson
>> >>
>> >> ==================================================
>> >> Open Grid Scheduler - The Official Open Source Grid Engine
>> >> http://gridscheduler.sourceforge.net/
>> >> http://gridscheduler.sourceforge.net/GridEngine/GridEngineCloud.html
>> >>
>> >>
>> >>
>> >> On Mon, Nov 18, 2013 at 8:29 AM, Sergio Mafra <sergiohmafra_at_gmail.com>
>> >> wrote:
>> >> > Hi everyone,
>> >> >
>> >> > I attended AWS Re:Invent this year and liked a lot. Right place to
>> >> > go.
>> >> > AWS announced there the new C3 family of HPC instances.
>> >> > Iīm anxious to test them in my models but I understand that
>> >> > StarCluster must
>> >> > take aware of them in the code.
>> >> > So when will be the C3s available?
>> >> >
>> >> > All the best,
>> >> >
>> >> > Sergio Mafra
>> >> >
>> >>
>> >> > _______________________________________________
>> >> > StarCluster mailing list
>> >> > StarCluster_at_mit.edu
>> >> > http://mailman.mit.edu/mailman/listinfo/starcluster
>> >> >
>> >>
>> >>
>> >>
>> >>
>
>
Received on Thu Jan 09 2014 - 20:25:28 EST