IETF
l2vpn
l2vpn@jabber.ietf.org
Thursday, 17 November 2011< ^ >
Room Configuration

GMT+0
[01:00:14] suz joins the room
[01:00:17] suz leaves the room
[01:01:16] hoyaj@jabber.org joins the room
[01:03:32] narten joins the room
[01:06:17] behcet.sarikaya joins the room
[01:09:40] adrianfarrel joins the room
[01:10:06] Bill joins the room
[01:11:04] marshall joins the room
[01:11:12] <marshall> Hello
[01:11:17] <marshall> I am going to jabber scribve
[01:11:40] <marshall> this meeting is an attempt to gather information about VPNs in the Data Center
[01:11:45] <marshall> Draft 1
[01:11:47] <behcet.sarikaya> Thanks Marshall!
[01:12:19] <marshall> Cloud Networking: Framework and VPN Applicability
http://tools.ietf.org/html/draft-bitar-datacenter-vpn-applicability-01
Nabil Bitar (nabil.n.bitar@verizon.com <mailto:nabil.n.bitar@verizon.com>)
[01:12:30] <Bill> I don't see these slides online.
[01:13:20] <marshall> Multi-tenant data center and data center interconnect requirements
[01:13:35] becarpenter joins the room
[01:13:42] Stewart Bryant joins the room
[01:14:31] <Stewart Bryant> l2vpn 8 or l2vpn 9 is the current slide deck
[01:14:56] <becarpenter> http://www.ietf.org/proceedings/82/slides/l2vpn-8.ppt
[01:15:51] <marshall> Thanks Brian
[01:16:04] Benson Schliesser joins the room
[01:16:48] <marshall> We want to support large scale DCs and optimize networkj resource utilization
[01:16:54] <marshall> slide 6
[01:17:08] <marshall> Also important is path mobilty
[01:17:21] <marshall> and VM mobility
[01:17:46] <marshall> reduction of packet loss during transit states (i.e., when VMs are moving)
[01:18:08] <marshall> not necessarily optimal routing, but something present during a transition
[01:18:46] <marshall> VM mobility, maintianing existing sessions, which usually means maintain MACs and IP addresses
[01:19:10] <marshall> optimal traffic forwarding during transition
[01:19:20] <marshall> page 7
[01:19:35] <marshall> auto-discovery by the network of
[01:19:44] <marshall> of a VM location, support for OAM
[01:19:49] Milo joins the room
[01:20:07] <marshall> also, need to ease the introduction of new DC networking tech in existing environments
[01:20:42] <marshall> David Darson : It would be helpful to specify whether these things are L2 or L3 or botg
[01:20:44] <marshall> both
[01:21:01] <Benson Schliesser> s/David Darson/Dave McDysan/
[01:21:04] <marshall> A : The draft talks about existing VPN solutions, layer 2 and layer 3
[01:21:16] <marshall> NEXT
[01:21:33] <Benson Schliesser> same preso - swap Nabil as speaker for Florin as speaker
[01:21:46] <marshall> Florin Balus
[01:22:13] <marshall> page 8 I believe
[01:22:32] <marshall> Layer 3 option - e.g. RFC 4364
[01:22:48] <marshall> Layer 2 options - VLAN, PBB, TRILL
[01:23:09] <marshall> in current draft PBB with L2 options have been covered
[01:23:20] <marshall> need to expand on IPsec part
[01:24:47] <marshall> Next slide - 10
[01:25:06] <marshall> PBB + L2VPN applicability for cloud networking
[01:25:18] <marshall> slide 11
[01:25:33] <marshall> supported tunneling options for 24b ISID tag over ethernet
[01:25:45] <marshall> 802.1ah-2008
[01:25:57] <marshall> ethernet over IP or MPLS tunneling
[01:26:12] <marshall> other more optimized tinneling options could be explored
[01:26:57] <marshall> there has to be a gateway function ot interconnect to existing customer sites
[01:27:02] <marshall> next sliude 12
[01:27:20] <marshall> 2 networks running ISID tags
[01:27:40] <marshall> DCGW tramslates back to regular VPLS
[01:27:43] <marshall> or
[01:27:49] <marshall> Slide 13
[01:28:13] <marshall> we have a use cae how VLAN and I-SID tags might interoperate
[01:28:18] <marshall> slide 14
[01:28:25] <marshall> control plan options
[01:28:57] <marshall> slide 15
[01:29:54] <marshall> you can have IP routing, IP routing with MPLS toolset, or legacy STP/MC-=LAG or ISIS (SPB)
[01:30:01] <marshall> slide 16
[01:30:26] <marshall> IS-IS BGP based control plain is a reuse of exisitng IETF technology
[01:30:44] <marshall> service autodiscovery is also important
[01:31:28] <marshall> We have to go 1 step further and go to Inter-AS, maybe with BGP
[01:31:33] <marshall> slide 17
[01:31:37] <marshall> other work
[01:31:59] <marshall> disccusion of VM Mobility, ARP supproession, ARP broadcast control
[01:32:01] <marshall> slide 18
[01:32:17] <marshall> table of component versus toolset used
[01:33:08] <marshall> slide 19
[01:33:13] <marshall> another tunnel
[01:33:24] <marshall> PBB+L2VPN challenges
[01:34:06] <marshall> another table, sorry
[01:34:24] <marshall> VM mobiity is a work in progress
[01:34:28] hoyaj@jabber.org leaves the room
[01:34:46] <marshall> multicast efficiency is a ?
[01:34:55] <marshall> [didn't catch that]
[01:34:59] <marshall> slude 20
[01:35:05] <marshall> Potential work items
[01:35:15] <marshall> IP tunneling optimization for ISID
[01:35:30] <marshall> network autoprovisiioning
[01:35:46] <marshall> broadcast and multicast over IP core requries work
[01:36:15] <marshall> also need a tunnel and service address translation between cloud provider and tenant provider
[01:36:41] <marshall> questions ?
[01:36:47] <marshall> I can relay for remote people
[01:37:27] <marshall> ? : Clarification : you touched on 802.1 but you didn't mention AQ
[01:37:30] <marshall> I meant AQ
[01:38:25] <marshall> ? : I likes the draft it was very good. For address relsolution, there is a big section. In ARMD we have worked on this - it is not a
[01:38:43] <marshall> big impact on servers - they have lots of CPUs.
[01:38:52] <marshall> The problems are in the routers
[01:39:06] <marshall> people there have done the work
[01:39:06] <Benson Schliesser> previous speaker at mic, might have been Dave Allen. Current speaker is Linda Dunbar.
[01:39:25] <marshall> thanks
[01:39:57] <marshall> Linde : another one is autoprovisioning - IEEE has PBB already, which should be used
[01:40:18] <marshall> Florin : We didn't want to describe solutions in this talk
[01:40:33] <marshall> Adrian Farrell ; thanks for putting this together
[01:40:51] <marshall> On this topic, you flip flop betwee cloud and data center
[01:40:55] <marshall> can you compare and contrast
[01:41:27] <marshall> Nabil: Cloud - set of data centers and the conections between them
[01:41:46] <marshall> Donald Eastlake - do you want to add TRILL ?
[01:42:05] <marshall> Florin : We wanted to put in places for people to add things
[01:42:10] <marshall> Donald : So, yes
[01:42:16] <marshall> Florin : Yes
[01:43:17] <marshall> Alley [?] : IP encapsulation means you can use multipoint to multipoint - a pseudowire carrys
[01:43:27] <Benson Schliesser> (Ali Sajassi)
[01:43:28] <marshall> I did a draft on M-VPLS
[01:43:36] <marshall> thanks
[01:44:06] <marshall> if you read that it is very similar to the VX standard you are mentioning
[01:44:26] <marshall> Nabil : the idea of this work was to use what exists
[01:44:58] <marshall> Kiri : I am going to attack you just because you jappen to be there
[01:45:11] <marshall> We have to stop ? at layer 2 and move to layer 3
[01:45:32] <marshall> Florin : There is a need for Layer 2 tunneling
[01:45:45] <Benson Schliesser> Nabil's comment was: Not "use what exists" but to say "these exist", not to pick one. We need to discuss the requirements. Thomas sets this up next.
[01:45:55] <Benson Schliesser> previous comment came from Kireeti Kompella.
[01:46:06] <Benson Schliesser> (s/Kiri)
[01:46:23] <Benson Schliesser> Question at Mic just came from Warren Kumari
[01:47:00] <marshall> Florin ; You are talking about using a flat layer 2 tunneling. It's a matter of whether you use Mac addresses or IP addresses. We are not against IP tnneling
[01:47:12] <marshall> NEXT : Thomas Narten
[01:47:18] <marshall> NVO3
[01:47:25] akatlas joins the room
[01:47:34] <marshall> Not my own work - 10 to 15 people behind the scenes
[01:47:39] <marshall> Slide 2
[01:47:41] <Bill> http://www.ietf.org/proceedings/82/slides/l2vpn-9.pdf
[01:47:42] <marshall> Level set
[01:47:59] <marshall> There are themes I want everyone to take away
[01:48:07] <marshall> focus on prioblems in the Data Center
[01:48:19] <marshall> NOT propose a specific solution
[01:48:36] <marshall> there is industry support already
[01:48:38] <marshall> we don't want to talk about prcoess
[01:48:45] <marshall> we'll sort that out later
[01:49:04] <marshall> industry is moving - we have a short window to engage
[01:49:12] <marshall> High Level
[01:49:19] <marshall> Imagine a Data Center
[01:49:28] <marshall> cloud provider maybe
[01:49:39] <marshall> multiple, non-trusting tenants
[01:50:02] <marshall> tenant wants to create a virtual netowork - network as a service
[01:50:10] <marshall> lots of ways to do this
[01:50:17] <marshall> slide 3
[01:50:21] <marshall> tenant requirementss
[01:50:31] <marshall> VMs think they are connected to a real netwotk
[01:50:56] <marshall> each VN (Virtual Network) uses its own address space
[01:51:04] <marshall> (that was VN 2 lines up)
[01:51:23] <marshall> VNs are fully isolated from each other
[01:52:01] <marshall> one tenant's traffic is contrained from another's except through well defined entry points
[01:52:07] <marshall> slide 4
[01:52:19] <marshall> logical view - VNs are fully isolated
[01:52:41] <marshall> could be connected to public internet, VPN out, etc.
[01:52:44] <marshall> slude 5
[01:52:55] <marshall> the data center provider has a different view
[01:53:40] <marshall> want ability to place VM's anywhere in the datacenter, without being constrained by the physical network
[01:54:03] <marshall> you can make all of this happen to ot everying in one L2 VLAN
[01:54:10] <marshall> this is not tenable any more
[01:54:18] <marshall> this problem will just get worse
[01:54:25] <marshall> no magic bullet here
[01:55:06] <marshall> two are basically in conflict - making L2VPNs larger runs into scaling problems
[01:55:19] <marshall> slude 6 - other requirements, from a DC perspective
[01:55:38] <marshall> today you have VLAN info tied to the VM
[01:55:48] <marshall> you can do this, but it is complex
[01:56:19] <marshall> [please, oh please, do not use VNs and VMs on the same slide!!!]
[01:57:02] <marshall> This is a _data center_ problem. VPNs are not driving this. DCs now may have or approach 1 million physical machines
[01:57:11] <marshall> slide 6 ?
[01:57:22] <marshall> physical and logical view
[01:58:03] <marshall> slide 7 : summary of requirements
[01:58:11] <marshall> multi-tenant support
[01:58:18] <marshall> support VM placement anywhere
[01:58:28] <marshall> on-demand elastic provision of resources
[01:58:37] <marshall> small forwarding tables in switches
[01:58:53] <marshall> decouple logical and physical network config
[01:59:18] <marshall> want to return to a model where switches do not know everything about the network
[01:59:31] <marshall> slide 7 ;
[01:59:37] <marshall> Data center today
[01:59:54] <marshall> servers < L2 VLANs < IP Cores
[02:00:09] <marshall> "<" meaning "included inside"
[02:00:35] <marshall> VPN comes in, terminates at a box, and goes into a VLAN
[02:00:57] <marshall> focus of approach is n the L2 VlLANs
[02:01:05] <marshall> NOVO3
[02:01:28] <marshall> new slide
[02:01:41] <marshall> what you see depends on where you look
[02:02:17] <marshall> NOT motivated by VPNs coming into theDC - but will connect to them
[02:02:38] <marshall> highly dynamic changes (network reconvirgues in subseconds)
[02:02:49] <marshall> approach is to use overlays
[02:02:59] <marshall> encapsulate with a SHIM header
[02:03:39] <marshall> overlay heafer will carry a Virttual Network Identifies - VNID
[02:04:06] <marshall> needs to be a "large enough" space - at least 24 bits - don't want to run out
[02:04:31] <marshall> switch takes packet, encapsulates it, and send it to far end
[02:04:36] <marshall> next slide
[02:05:00] <marshall> communication between VM1 and VM3
[02:05:32] <marshall> Hypervisor encapsulates and tunnels packet tonext hypervisor
[02:06:12] <marshall> How does hypervisor know how to get to the other end and where the otehr hypervisor is
[02:06:17] <marshall> next sldie
[02:06:39] <marshall> multiple encapsulations are already used
[02:06:50] <marshall> control plane is the interesting placeDimitri
[02:07:21] <marshall> there needs to be gateways
[02:08:15] <marshall> Thomas : Having multiple encappsulations would be be a problem if you didn't know which one to use
[02:08:55] <marshall> Thomas : let's take this up at the end - we may cover this
[02:09:34] <marshall> ? : Need to make sure that a single communication error doesn't cause problems
[02:09:39] <marshall> Thomas :
[02:09:42] <marshall> Next slide
[02:09:49] <marshall> In terms of the control plane
[02:09:57] <Benson Schliesser> previous comment at mic came from Dave McDyson
[02:10:02] <marshall> need mechanisms to
[02:10:11] <marshall> -populate mapping tables
[02:10:22] <marshall> deliver multi--destination frames
[02:10:57] <marshall> - registration for endpoints to tell switches when the EP joins a VN, etc.
[02:11:13] <marshall> next slide
[02:11:13] <marshall> 2 address mapping approaches
[02:11:38] <marshall> resuse control plane from IEEE 802.1 bridging
[02:11:55] <marshall> flood packets with unknwon unicast addresses, maybe using multicast
[02:12:13] <marshall> simple approach that doesn't scale well
[02:12:26] <Benson Schliesser> clarifying question from Ali Sajassi
[02:12:44] <marshall> Clarificaition question : If the learning s done in the data plane, isn't it a control plane
[02:12:53] <marshall> Thoma : I use the term losely
[02:13:28] <marshall> ? : Your viewgraph shows that unicast and multicast paths are not the same, which might be an issue
[02:13:33] duan.chen joins the room
[02:13:39] <marshall> next slide
[02:13:46] duan.chen leaves the room
[02:14:11] <marshall> address mapping - directory service
[02:14:31] <marshall> use a centralized directory to story address mappings
[02:14:53] duan.chen joins the room
[02:15:11] <marshall> centralized is a minnomer - need replication and backing
[02:15:15] duan.chen leaves the room
[02:15:25] <marshall> this is a high level approach
[02:15:38] <marshall> a learning based solution is not sufficient for all deployments
[02:15:59] <marshall> ? : Most people are going to a push approach
[02:16:15] <Bill> Igor Gashinsky was talking about the query-based vs. push approach
[02:16:21] <marshall> the central system will know where things need to go
[02:16:27] <marshall> so, a push approach
[02:16:30] <marshall> next slide
[02:16:34] <marshall> the big picture
[02:16:45] <marshall> NV03 doesn't do everything
[02:17:35] <marshall> orchestration systems that people are implementing are different
[02:17:38] <marshall> next slide
[02:17:55] <marshall> 2 exisitng proposals for implementing overlays
[02:18:02] <marshall> VXLAN and NVGRE
[02:18:17] <marshall> short window of opportunity for the IETF to get involved
[02:18:22] <marshall> relaed work
[02:18:32] <marshall> next slide
[02:18:40] <marshall> TRILL is L2
[02:18:56] <marshall> Some people want a L3 solution
[02:19:24] <marshall> IEEE has Shortest Path Bridging (SPB)
[02:20:35] <marshall> my impression is the L2VPN is fundamentally about using SPs to glue togethers VPNs across a network
[02:20:50] <marshall> L2VPN will NOT be the only approach
[02:21:11] <marshall> next slide\
[02:21:16] <marshall> L2 vs L3 VPNs
[02:21:34] <marshall> NVO3 is NOT just L2 over L3
[02:21:46] <marshall> control plane issues are agnostri
[02:22:39] <marshall> next slide
[02:22:47] <marshall> NVO3 and VPNs
[02:23:16] <marshall> will still need to connect to VPNs, but that is secondary
[02:23:36] <marshall> cost benefit of verylaying is extremely attractive to
[02:23:42] <marshall> operators
[02:24:25] <marshall> TRILL, SPB and NVO3
[02:24:44] <marshall> Again, the IETF has a fairly short window to act
[02:24:55] <marshall> end
[02:25:44] <marshall> now is open mike
[02:25:52] <marshall> again, I can replay questions
[02:26:46] <marshall> ? You want to have a VM, some ID and a map to the network
[02:26:49] <Bill> Ping Pan
[02:27:19] <marshall> orchestration is part of the SDN BOF this afternoon and is part of IETF work
[02:27:32] <marshall> John Scutter : there is a lot to unpact
[02:27:53] <marshall> you hit a bullet point that said hurry hurry hurry
[02:28:05] <marshall> but short cuts leads to long delays
[02:28:25] <marshall> there was a lot of wheel re-inventing in there
[02:28:45] <marshall> Thomas : I am not trying to rush this through
[02:30:11] Benson Schliesser leaves the room
[02:30:24] Benson Schliesser joins the room
[02:30:56] <marshall> If you look at the overall high level vision there are lots of peices. Not all need standardization
[02:31:14] <marshall> Q : We need to clarify what we mean by orchestration
[02:31:32] duan.chen joins the room
[02:31:39] <marshall> Second point - Sumamry points
[02:31:54] <marshall> you seem to say that NVO3 is very different
[02:32:09] <marshall> but it's not really any different
[02:32:39] <marshall> Thomas : At one level I would agree with you
[02:33:04] <marshall> Q : We need some overlay model and some way to tunnel throgh
[02:33:23] duan.chen leaves the room
[02:33:30] <marshall> Eric : You talked about sclaing itup, but I wonder if we shouldn't be thinking about scaling it down
[02:34:29] <marshall> the other thing, are you thinking of having multiple control plans over multuple encapsultaions
[02:34:30] hoyaj@jabber.org joins the room
[02:34:48] <marshall> at the low scale, you do learning
[02:35:02] <marshall> at the high scale you set up a database
[02:37:44] <marshall> Stewarrt : I detect we are drifting into the general discussion. Could anyone who is going to ask a general question should go to the end of the line
[02:41:24] <marshall> Thomas : Hearing from DC operators, the approaches that L2VPN are doing do not meet their needs
[02:42:33] <marshall> Thomas Morin : How do you envision a set of VMs connected to the outside world ?
[02:42:46] <marshall> Thomas Narten : Something similar to current VLANs
[02:43:26] <marshall> Thomas M : A specific product ?
[02:43:43] <marshall> Thomas N : That is not for the IETF to decide
[02:44:11] <marshall> Alex : There needs to be a gateway which needs to be compatible with the outside world
[02:48:51] <marshall> Martin from Juniper - I like a lot of the comments, references to the scalablity to millions of hosts, only storing infrastructure state
[02:49:16] <marshall> The technology that can handle this is IP, so more iP please
[02:50:24] <marshall> Paul Hagen : Difference perspectives - what is a data center operator
[02:51:36] <marshall> Chris (previously at Telstra) : I will have to agree with Igor, this is clear and concise and i think this is something we need to do
[02:52:32] <marshall> If I look at the data center and the number of gateways, I would rather force a change in the few gateways rather than the 10's of 1000's of network devices inside the data center
[02:53:44] <marshall> Stewart : We have 3 questions
[02:53:47] <marshall> What is the problem?
[02:53:54] <marshall> What new work has to be done?
[02:54:04] <marshall> Do we agree that the work has to be done
[02:54:06] <marshall> ?
[02:54:41] <marshall> Florin : It's not just about L3 over L2 but L3 over L3. We also need work over the IP core
[02:55:10] <marshall> Dave Allen : Two ways forward, a bridging overlay and a ? restructure
[02:55:31] <marshall> if the goal is to reinvent here what has been done elsewhere, that is not a help
[02:55:52] <marshall> ? : It is obvious that we have a problem to make things scale
[02:57:36] <marshall> Thomas Morin : The problem statement is good but it is too early to decide question of connectivity
[02:57:51] <marshall> We must be casreful about not reinventing the wheel
[02:58:44] <marshall> there are people who want to to do things in hypervisors and people who ? [couldn't catch that one]
[02:59:02] <marshall> I think we need to allow for both solutons to be used by an operator
[02:59:30] <marshall> We must also be careful of something that is like blackmail
[03:00:01] <marshall> Ping : VXLAN is a real solution that will go into the REDHAT kernel soon
[03:00:17] <marshall> but in my mind this is a mangement problem
[03:00:46] <marshall> We should have an interim meeting
[03:01:25] <marshall> Stewart : I think we need to get through this, and then take some straw polls\
[03:01:59] <marshall> Eric {?} : First we have to understand the VN requirements independent of technologies
[03:03:51] akatlas leaves the room
[03:05:29] hoyaj@jabber.org leaves the room
[03:06:01] <marshall> Adrian : Not wearing hats - thanks for the people who worked on this
[03:08:46] <marshall> Eric : Answering the questions Yes, yes and yes
[03:08:59] adrianfarrel leaves the room
[03:09:28] <marshall> think about IP over IP
[03:10:20] <marshall> Dino F : Absolutely the IETF needs to work on this - we need to create less options and standardized solutions
[03:10:29] <marshall> one and one encapsulation is importnat
[03:10:36] <marshall> Chris, RedHat :
[03:11:10] <marshall> New work - I beleive that the framing format is not as critical as the control plane
[03:11:35] <marshall> Igor : This is awesum work
[03:11:52] <marshall> the control plane, the mapping plane is what matters
[03:12:34] <marshall> Manuel from DT : There is quite a complementaty set of requirements from today and yesterday
[03:13:56] <marshall> let
[03:14:08] <marshall> lets do some hands
[03:14:29] <marshall> how many people who have read the problem statement ?
[03:14:32] <marshall> lots
[03:14:44] <marshall> how feels it is a good starting point
[03:15:06] <marshall> mixed
[03:15:35] <marshall> Stewart : Who thinks we needs more data center input for this
[03:15:45] <marshall> about /12 yes and 1/2 no
[03:16:11] <marshall> voice : I don't think that these are mutually exclusive choices
[03:16:33] <marshall> Comment: I thnk that the IETF needs more DataCenter clue
[03:16:40] <marshall> [many hands on that]
[03:17:25] <Bill> "Comment": was George Swallow
[03:17:29] Milo leaves the room
[03:17:37] <marshall> Thomas Narten : I am always in favor of minimal, but what is minimal
[03:17:38] <Bill> "voice" was Lou Berger
[03:18:11] <marshall> Who feels we should drive towards one and only one control planes?
[03:18:19] <marshall> a handful of hands
[03:19:47] <marshall> Nabil : We need to be clear about what 1 control plane means
[03:20:01] <marshall> Pat : I think that control plane is a tool box
[03:20:33] becarpenter leaves the room: offline
[03:20:36] <marshall> that may mean more than one tool, but a minimal set
[03:20:55] <marshall> Lou Berger : Perjhaps the questions are at too low a level
[03:21:02] <marshall> we need to figure out the framework yet
[03:22:05] Yuri Nawata joins the room
[03:22:59] Yuri Nawata leaves the room
[03:24:06] <marshall> Stwart : Logistics questions
[03:24:45] <marshall> Do we need a interim before Paris ?
[03:24:55] <marshall> medium number yes or no
[03:25:09] <marshall> this will need hosts
[03:25:32] <marshall> there is a meeting on SDN this afternoon that will gather more data
[03:26:15] <marshall> Stewart / Thomas : It would be an interim to deal with the larger data center problem ?
[03:27:00] narten leaves the room
[03:27:04] <marshall> substantially more hands yes than no
[03:27:56] <marshall> We need a DC mailing list (not VPN4DC or NVO3 lists)
[03:28:29] Benson Schliesser leaves the room
[03:28:35] Stewart Bryant leaves the room
[03:29:23] adrianfarrel joins the room
[03:29:38] behcet.sarikaya leaves the room
[03:31:05] marshall leaves the room
[03:33:55] Bill leaves the room: Computer went to sleep
[03:54:33] Bill joins the room
[03:55:01] adrianfarrel leaves the room
[03:55:46] Bill leaves the room
[04:42:38] Milo joins the room
[05:34:46] Stewart Bryant joins the room
[05:56:00] Stewart Bryant leaves the room
[06:39:41] Milo leaves the room
[07:22:28] Stewart Bryant joins the room
[09:28:49] Stewart Bryant leaves the room
[10:01:23] Stewart Bryant joins the room
[11:49:42] Stewart Bryant leaves the room
[12:33:11] Stewart Bryant joins the room
[13:02:14] Stewart Bryant leaves the room
[13:33:28] Stewart Bryant joins the room
[13:33:41] Stewart Bryant leaves the room
[17:13:58] Stewart Bryant joins the room
[18:31:12] Stewart Bryant leaves the room