Re: [Tsv-art] Tsvart last call review of draft-ietf-nvo3-vmm-04

"Black, David" <> Fri, 14 September 2018 01:03 UTC

Return-Path: <>
Received: from localhost (localhost []) by (Postfix) with ESMTP id 5C5B8127AC2; Thu, 13 Sep 2018 18:03:34 -0700 (PDT)
X-Virus-Scanned: amavisd-new at
X-Spam-Flag: NO
X-Spam-Score: -2.71
X-Spam-Status: No, score=-2.71 tagged_above=-999 required=5 tests=[BAYES_00=-1.9, DKIM_SIGNED=0.1, DKIM_VALID=-0.1, DKIM_VALID_AU=-0.1, HTML_MESSAGE=0.001, RCVD_IN_DNSWL_LOW=-0.7, SPF_PASS=-0.001, T_DKIMWL_WL_HIGH=-0.01] autolearn=ham autolearn_force=no
Authentication-Results: (amavisd-new); dkim=pass (1024-bit key) header.b=WA3cv2jO; dkim=fail (1024-bit key) reason="fail (message has been altered)" header.b=itMUruQU
Received: from ([]) by localhost ( []) (amavisd-new, port 10024) with ESMTP id rNXc37gEE9Lh; Thu, 13 Sep 2018 18:03:29 -0700 (PDT)
Received: from ( []) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by (Postfix) with ESMTPS id 38BF21277D2; Thu, 13 Sep 2018 18:03:26 -0700 (PDT)
DKIM-Signature: v=1; a=rsa-sha256; c=simple/simple;;; q=dns/txt; s=smtpout; t=1536887009; x=1568423009; h=from:cc:to:subject:date:message-id:references: in-reply-to:mime-version; bh=ZbsyjH4MgV3/nQYl0IywFBj1XKqvu/9sBLtzb8k2UOs=; b=WA3cv2jOoWTzwA35wziUkcVnH7R/55kI7acXeXEnCZ5yDnN1jfK7vanP 5OAwQBMpkmgD3FDM+ygG43RFQ/HsQIWF89NaOeoChF8qnEs6AoUBUvaI8 SFD54QkRoHSI2bI5h3iBMrdUlqHACtbcwbxCZJmWr+eFv9+X6Na+r70Bx 8=;
X-IronPort-Anti-Spam-Filtered: true
Received: from ([]) by with ESMTP/TLS/AES256-SHA256; 13 Sep 2018 20:03:24 -0500
Received: from pps.filterd ( []) by ( with SMTP id w8E0wEED147062; Thu, 13 Sep 2018 21:03:24 -0400
Received: from ( []) by with ESMTP id 2mg0h0ghff-1 (version=TLSv1.2 cipher=ECDHE-RSA-AES256-GCM-SHA384 bits=256 verify=FAIL); Thu, 13 Sep 2018 21:03:23 -0400
From: "Black, David" <>
Cc: "" <>, "" <>, "" <>, "Black, David" <>
Received: from ([]) by with ESMTP/TLS/DHE-RSA-AES256-SHA256; 14 Sep 2018 07:03:07 +0600
Received: from ( []) by (Sentrion-MTA-4.3.1/Sentrion-MTA-4.3.0) with ESMTP id w8E13I5u017633 (version=TLSv1.2 cipher=ECDHE-RSA-AES256-GCM-SHA384 bits=256 verify=NO); Thu, 13 Sep 2018 21:03:19 -0400
X-DKIM: OpenDKIM Filter v2.4.3 w8E13I5u017633
DKIM-Signature: v=1; a=rsa-sha1; c=relaxed/relaxed;; s=jan2013; t=1536887001; bh=fcRNNRxptuTxL9xbnSb2+94VWI0=; h=From:To:CC:Subject:Date:Message-ID:References:In-Reply-To: Content-Type:MIME-Version; b=itMUruQUAHZJfPCEIv2pU7iE2yl+NVcnYR5z2DyMaMbuNqrN5EnJgaYVfG8nUyTx5 tYbhR+ngp/WFy9F++Jp89CnxJWI8LIBQ9Tw/Alzh7OILHHN05jHT+M0IxVvKTAGX5W yntQoWm1A1bTKCXXi9Z/Z63SP/8+vA/K0Xn8giL8=
X-DKIM: OpenDKIM Filter v2.4.3 w8E13I5u017633
Received: from ( []) by (RSA Interceptor); Thu, 13 Sep 2018 21:02:54 -0400
Received: from ( []) by (Sentrion-MTA-4.3.1/Sentrion-MTA-4.3.0) with ESMTP id w8E12sKK003853 (version=TLSv1.2 cipher=ECDHE-RSA-AES256-SHA384 bits=256 verify=FAIL); Thu, 13 Sep 2018 21:02:55 -0400
Received: from ([fe80::849f:5da2:11b:4385]) by ([]) with mapi id 14.03.0399.000; Thu, 13 Sep 2018 21:02:54 -0400
To: Linda Dunbar <>, Bob Briscoe <>, "" <>
Thread-Topic: Tsvart last call review of draft-ietf-nvo3-vmm-04
Thread-Index: AQHUSTKisXCjryIZFECZqK2mckAgn6Tu+cXw
Date: Fri, 14 Sep 2018 01:02:54 +0000
Message-ID: <>
References: <> <>
In-Reply-To: <>
Accept-Language: en-US
Content-Language: en-US
x-originating-ip: []
Content-Type: multipart/alternative; boundary="_000_CE03DB3D7B45C245BCA0D2432779493630287A13MX307CL04corpem_"
MIME-Version: 1.0
X-RSA-Classifications: public
X-Proofpoint-Virus-Version: vendor=fsecure engine=2.50.10434:, , definitions=2018-09-13_15:, , signatures=0
X-Proofpoint-Spam-Details: rule=outbound_notspam policy=outbound score=0 priorityscore=1501 malwarescore=0 suspectscore=0 phishscore=0 bulkscore=0 spamscore=0 clxscore=1011 lowpriorityscore=0 mlxscore=0 impostorscore=0 mlxlogscore=999 adultscore=0 classifier=spam adjust=0 reason=mlx scancount=1 engine=8.0.1-1807170000 definitions=main-1809140009
Archived-At: <>
Subject: Re: [Tsv-art] Tsvart last call review of draft-ietf-nvo3-vmm-04
X-Mailman-Version: 2.1.29
Precedence: list
List-Id: Transport Area Review Team <>
List-Unsubscribe: <>, <>
List-Archive: <>
List-Post: <>
List-Help: <>
List-Subscribe: <>, <>
X-List-Received-Date: Fri, 14 Sep 2018 01:03:35 -0000

> #. It does not seem as if the NVO WG has discussed the purpose of using normative text in this draft. See detailed comments.

> [Linda] The “Intended status” of the draft is “Best Current Practice”. So all the text are not “normative”. Is it Okay?

Not really – this draft might be better targeted as “Informational” as it is not a comprehensive review of current practice (best or otherwise) nor an overall set of recommendations, e.g., as Bob wrote “it just asserts what appears to be one view of how a whole VM Mobility system works.”

Thanks, --David

From: Tsv-art [] On Behalf Of Linda Dunbar
Sent: Monday, September 10, 2018 2:17 PM
To: Bob Briscoe;
Subject: Re: [Tsv-art] Tsvart last call review of draft-ietf-nvo3-vmm-04


Thank you very much for reviewing the draft and provided in-depth comments. I am very sorry for the delayed response due to traveling.

Replies to your comments are inserted below marked by [Linda]:

-----Original Message-----
From: Bob Briscoe []
Sent: Monday, September 03, 2018 9:45 PM
Subject: Tsvart last call review of draft-ietf-nvo3-vmm-04

Reviewer: Bob Briscoe
Review result: Not Ready

I have been selected as the Transport Directorate reviewer for this draft. The Transport Directorate seeks to review all transport or transport-related drafts as they pass through IETF last call and IESG review, and sometimes on special request. The purpose of the review is to provide assistance to the Transport ADs. For more information about the Transport Directorate Reviews and the Transport Area Review Team, please see ​

In this case, very very few of the review comments relate to transport issues, although the greatest issue concerns a desire that the network could pause or stop connections during L3 VM Mobility, which is certainly a transport issue.

[Linda] There is “Hot Migration” with transport service continuing, and there is a “Cold Migration”, which is a common practice in many data centers, which stop the task running on the old place and move to the new place before restart as described in the Task Migration.
Is it helpful to add this description to the draft?


The technical aspects of the draft concerning L2 VM mobility (within a subnet) seem sound. However, this is only part of the draft, which has the following

#. The introduction does not say what the purpose of publishing this draft is.
It seems that, rather than describing a specific protocol or protocols, it intends to describe the overall system procedure that would typically be used in DCs for VM mobility. It is tagged as a BCP, but it does not say who needs this BCP, why it is useful for the IETF to publish this BCP, how wide the authors' knowledge is of current practice (given DCs are private), or why this is a BCP rather than a protocol spec.

[Linda] The first paragraph on Page 3 has the description why VM Mobility is needed. Is it helpful to move this paragraph to the beginning of the Introduction Section?
“Virtualization which is being used in almost all of today’s data
centers enables many virtual machines to run on a single physical
computer or compute server. Virtual machines (VM) need hypervisor
running on the physical compute server to provide them shared
processor/memory/storage. Network connectivity is provided by the
network virtualization edge (NVE) [RFC8014]. Being able to move VMs
dynamically, or live migration, from one server to another allows for
dynamic load balancing or work distribution and thus it is a highly
desirable feature [RFC7364].”

The draft starts out (S.3) as if it intends to say what a good VM Mobility protocol should or shouldn't do, but the rest of the document doesn't give any reasoning for these recommendations, it just asserts what appears to be one view of how a whole VM Mobility system works, sometimes referring to one example protocol RFC for a component part, but more often with no references or details.

[Linda] Is it helpful to move the paragraph above to the beginning of the Introduction Section? So that audience is aware of why VM Mobility is needed. And then follow up with what a good VM Mobility protocol should or shouldn't do?

#. It does not seem as if the NVO WG has discussed the purpose of using normative text in this draft. See detailed comments.

[Linda] The “Intended status” of the draft is “Best Current Practice”. So all the text are not “normative”. Is it Okay?

#. The draft silently slips back and forth between VM mobility and VM redundancy, without recognizing the differences. See detailed comments.

[Linda] There is only one usage of “redundancy” in the entire document, used under the context of “Hot standby option”, indicating  the “redundancy” of “the VMs in both primary and secondary domains have identical information and can provide services simultaneously as in load-share mode of operation” being expensive.

#. Please adopt different terminology than "source NVE" and "destination NVE", which are really poor choices of terms for an intermediate node. See detailed comments. Why not use "old NVE" and "new NVE", which is what you mean?
[Linda] Thanks for the suggestion. We will change to “Old NVE”, and “new NVE”.

#. Applicability is fairly clearly outlined, but it is not clear whether hosts corresponding with the mobile VMs are part of the same controlled environment or on the uncontrolled public Internet. See detailed comments.
[Linda] “Hosts” are the App running on the VM. It is the under the same controlled environment. Not on uncontrolled public internet.

#. Section 4.2.1 on L3 VM mobility reads like some potential half-thought-through ideas on how to solve L3 mobility, rather than current practice, let alone best current practice. Either current practice should be described instead, or the scope of the draft should be narrowed solely to L2 VM mobility. See detailed comments.
[Linda] This is refereeing to “Cold Migration”, which is a common practice in many data centers.

# The VM's file system is described as state that moves with the VM (S.6), but VM mobility solutions often move the VM but stitch it back to its (unmoved) storage. Conversely, the storage can also move independent of the VM.
[Linda] It depends. When a VM move to a different zone, the storage/file can becomes inaccessible.

#. The draft omits some of the security, transport and management aspects of VM mobility. See detailed comments.
[Linda] Can you provide some text?

#. The draft reads as if different sections have been written by different authors and no-one has edited the whole to give it a coherent structure, or to ensure consistency (both technical and editorial) between the parts. See detailed comments.

[Linda] we can improve.

#. The quality of the English grammar does not allow a reviewer to concentrate on the technical aspects rather than the English. It would have been useful if one of the English-speaking co-authors had improved the English before submission for review. See detailed comments.
[Linda] can you help?  Becoming a co-author to improve?

==Detailed Comments==

===#. Normative statements===

In the body of the document, there is just one occurrence of normative text (actually two "MUST"s, but both state a common requirement - just written separately for IPv4 and IPv6). This merely serves to imply that everything else the document says is less important or optional, which was probably not the intention.
[Linda] The goal is to indicate any solution in moving the VM “MUST” follow this rule. They make sense, aren’t they?

At the start there is a requirements section, which states what a VM Mobility protocol "SHOULD" or "SHOULD NOT" do. I think this is intended as a set of goals for the rest of the document. If so, these "SHOULDs" are not intended to apply to implementations, so they ought not to be capitalized.

[Linda] okay, will change.

The first requirement, "Data center network SHOULD support virtual machine mobility in IPv6", is written as a requirement on all DC networks, not on implementations. I assume this was intended to read as "Data center network virtual machine mobility protocols SHOULD support IPv6". Even then, it doesn't really add anything to say VM mobility should support v6 and it should support v4. A L2 solution won't. While undoubtedly, a L3 solution will at least support one of them.
[Linda]Agree. Will change it to “Data center that support IPv6 address should …”

I'm not sure that 'protocol' is the right word anyway; I think 'VM Mobility procedure' would be a better phrase, because it includes steps such as suspending the VM, which is more than a protocol.
[Linda] yes. Will change to “Procedure”.

The requirement "Virtual machine mobility protocol MAY support host routes to accomplish virtualization", is not followed up at all in the rest of the draft.
Even if this requirement stays, the last 3 words should be deleted.

[Linda] will change to “Host Route can be used to support the Virtual Machine Mobility Procedure.”

By the end of the draft, the solution falls far short of the most relevant "Requirements" anyway, so one assumes the title of the section ought to have been "Goals". Specifically, even in the simpler case of L2 VM mobility, S.4.1 says that triangular routing and tunnelling persist "until a neighbour cache entry times out". A cache timeout is about 10 orders of magnitude longer than the requirement to only persist "while handling packets in flight", which would be a few milliseconds at most (the time for packets to clear the network that were already launched into flight when the old VM stopped).

Whatever, it would be preferable for the draft to give rationale for these requirements, rather than just assert them. This would help to shed light on the merits of the different trade offs that solutions choose.

[Linda] Agree, will add.

===#. Mobility vs. Redundancy===

Redundancy and mobility have a lot of similarities, but they have different goals. With mobility, it is necessary to know the exact instant when one set of state is identical to the other so it can hand over. With redundancy, the aim is to keep two (or more) sets of state evolving through the same sequence of changes, but there is no need to know the point at which one is the same as the other was at a certain point.
[Linda] Agree with what you said. There is only one usage of “redundancy” in the entire document, used under the context of “Hot standby option”, indicating  the “redundancy” of  “the VMs in both primary and secondary domains have identical information and can provide services simultaneously as in load-share mode of operation” being expensive.

The draft slips from mobility to resilience in the following places:
* S.2. Terminology: Warm VM Mobility is defined without any ending, as if it is permanent replication. * S.7. "Handling of Hot, Warm and Cold Virtual Machine Mobility" is actually all about redundancy, and doesn't address mobility explicitly.

[Linda] Will add the definition “Hot Migration”, “cold migration”, and “warm migration”.

===#. Terminology===

Packets run from the source at A to the destination at B via NVE1, then via NVE2. Please don't call NVE1 and NVE2 the source NVE and the destination NVE.
In future, no-one will thank you for the apparent contradictions when they continually stumble over phrases like this one in S.4.1: "...send their packets to the source NVE".

The term "packets in flight" is used incorrectly to refer to all the packets sent to the old NVE after the VM has moved, even if they were launched into flight long after the old VM stopped receiving packets.

[Linda] thank for the comments. Will change.

BTW, I think s/before/after/ in: "that have old ARP or neighbor cache entry before VM or task migration".

I think: s/IP-based VM mobility/L3 VM mobility/ throughout, because "based"
sounds (to me) like the mobility control protocol is over (i.e. based on) IP.

===#. Applicability===

In section 4.2 it says that the protocol mostly used as the IP based task migration protocol is ILA. This implies that all hosts corresponding with the mobile VMs are either part of the same controlled environment, or they are proxied via nodes that are part of the same controlled environment (I only have passing knowledge of ILA, but I understand that it depends on ILA routers on the path). If I am correct, this aspect of scope needs to be made clear from the start.

Also under the heading of applicabiliy, the sentence "Since migrations should be relatively rare events" appears very late in the document (S.4.2.1). The assumed level of churn ought to be stated nearer the start.

[Linda] yes, under the same controlled environment.

===#. L3 Mobility===
L2 VM mobility is independent of the application, because resolution of L2 mappings is delegated to the stack. In contrast, L3 VM mobility is only feasible under certain conditions, because an application needs an IP address to open a socket (resolution of DNS names is not delegated to the stack, and apps can use IP addresses directly anyway).

Examples of the 'certain conditions':
a) /All/ applications used in the whole DC load balancing scheme contain IP address migration logic for /all/ their connections; b) VMs running solely applications that support IP address migration register this fact with the NVA, and it only select such VMs for mobility. c) An abstraction is layered over /all/ the IP addresses exposed to applications (at both ends) so that the IP addresses that applications use are solely identifiers  (e.g. ILA, LISP, HIP), not also locators.

The introduction says the draft is about VM mobility in a multi-tenant DC, so the DC admin will not know the range of applications being used. This excludes condition (a) above. When the draft says "...if all applications running are known to handle this gracefully...", it doesn't quantify just how restrictive this condition is, and it gives no explanation of how this knowledge might be 'known' or which function within the system 'knows' it.

S.4.2.1 contains what seems like plenty of arm-waving.
* "TCP connections could be automatically closed in the network stack during a migration event."
        o There is no TCP connection state in the network stack.
        o Even if the network starts to drop every packet, the TCP connection
        state persists in the end-points for a duration of the order of 30-90
        minutes (OS-dependent) before TCP deems the connection is broken. o
        Other transport protocols have similar designs (including the app-layer
        of protocols over UDP).
* "More involved approach to connection migration":
        o pausing the connection [does this refer to an actual feature of any
        L4 protocol?] o packaging connection state and sending to target [does
        this assume logic written into the application, or is this assuming the
        stack handles this and the app is restricted to using some form of
        separate identifier/locator addresses?] o instantiating connection
        state in the peer stack [ditto?].

There's some arm-waving in S.7 too:
  "Cold Virtual Machine mobility is facilitated by the VM initially
   sending an ARP or Neighbor Discovery message at the destination NVE
   but the source NVE not receiving any packets inflight."
   [How is it arranged for the source NVE not to receive any packets in flight?]

And in S.7:
  "In hot
   standby option, regarding TCP connections, one option is to start
   with and maintain TCP connections to two different VMs at the same
   [This sounds like resilience logic has been written into the application,
   which would be a special case but not something VM mobility infrastructure
   could depend on.]

[Linda] will add.

===#. Gaps===
#. Security Considerations: repeats issues in other drafts that are not specific to mobility, but it does not mention any security issues specifically due to VM mobility. It says that address spoofing may arise in a DC (sort-of implying it is worse than in non-DC environments, but not saying why). The handshake at the start of a connection (e.g. TCP, SCTP, QUIC) checks for source address spoofing. So L3 VM mobility would be more vulnerable to source address spoofing in cases where the mobile VM was the connection initiator and there was not a new handshake after the move. However, this draft does not contain any detailed mobility protocols, so it is not possible to identify any specific security flaws.

#. Transport Issues: Effect of delay on the transport: Cold mobility introduces significant delay, and other forms less, but still some delay. It should be pointed out that some applications (e.g. real-time) will therefore not be useful if subjected to VM mobility. Similarly, even a short period of delay will drive most congestion controls to severely reduce throughput. These points might be self-evident, but perhaps they should be stated explicitly.

BTW, in the L3 VM mobility case, the draft often refers to TCP connections, but the address bindings of any transport protocols would have to be migrated due to VM mobility (e.g. SCTP; sequences of datagrams over UDP; streams over UDP such as with RTP, QUIC).

#. Management Issues: perhaps the draft ought to recommend statistics gathering (e.g. time taken, amount of duplicate data) to aid a DC's future decisions on the cost-benefit of moving a VM. The OPSDIR review says a BCP does not /have/ to describe management issues, but this document seems to describe a whole system procedure, not just a protocol, which then surely includes the management plane.

[Linda] can you become a co-author and add those in?

===#. Incoherent Structure===

S.4.1. happens to talk about VMs moving, while S.4.2. happens to talk about tasks moving, but this is not the distinguishing aspect of these two sections (anyway, S.2. says "the draft uses task and VM interchangeably"): * "4.1 VM Migration" is about "L2 VM Mobility" so this ought to be the section heading, *
"4.2 Task Migration" is about "L3 VM Mobility" so this ought to be the section heading. It would also help not to switch from VM to task across these sections
- it's just a distraction.

S.4.1 needs better signposting of where each sub-case ends (Subsections might be useful to solve this): * IPv4 * end-user client * 2 paras starting "All NVEs communicating with this virtual machine..." [Not clear that the end-user case has ended and we have returned to the general IPv4 case?] * IPv6 [Strictly, it still hasn't said whether the end-user client case has ended.] [Also, it doesn't explain why there is no need for an end-user client case under IPv6?] Sections 5 & 6 seem to be about either L2 or L3 mobility, whereas Sections 7 &
8 seem to be restricted to L2.

The draft vacillates over what to do with packets arriving at the old NVE in the L3 case (see also L3 mobility above): * S4.2 first says packets are dropped, possibly with an ICMP error message;
  o then later it says they are silently dropped;
  o then in the very next sentence it says either silently drop them or forward
  them to the new location
* S.5 says they should not be lost, but instead delivered to the destination hypervisor
  o then it describes how they are tunnelled (which is not the same as

The order in which all the stages of mobilty are given is jumbled up across sections that also appear in arbitrary order: * S.5 prepares, establishes uses then stops a tunnel, but it doesn't say where the other stages fit between these steps
        o When tunneling packets, it talks about the *migrating* VM not the
        *migrated* VM, which implies tunnelling has started before the new VM
        is running. Does this imply there is a huge buffer? o It says "Stop
        Tunneling Packets - When source NVE stops receiving packets destined
        to..." but it is never clear when a source has stopped sending packets
        to a destination, unless it explicitly closes the connection (e.g. with
        a FIN in the case of TCP). Often there are long gaps between packets,
        because many flows are 'thin' (meaning the application frequently has
        nothing to send). These gaps can last for milliseconds, hours or even
        days without any implication that the connection has ended.
* Then S.6. describes moving state, but doesn't say that this is not after the previous tunnelling steps (or where it fits within those steps). * Then S.7 describes hot, warm and cold mobility, but doesn't lay out the tunnelling or steps to move state in each case. * Then S.8 says it's about VM life-cycle, but just gives the very first 3 steps for allocation of resources to a VM, then abruptly ends, without even starting the VM, let alone getting to move it.

S.5 exhibits another inconsistency by talking about the hypervisor, not the NVE.

==#. Nits==

Nits with the English are too numerous to mention them all. Below are pointers to general problems as well as some individual instances.

  "Layer 2 and Layer 3 protocols are described next.  In the following
   sections, we examine more advanced features."

Expand WSC, MSC and NVA on first use.

s/the VM moves in the same link/the VM moves in the same subnet/

"i.e. end-user clients ask for the same MAC address upon migration. [...] to ensure that the same IPv4 address is assigned to the VM." I think s/IPv4/MAC/ was intended?

"  All NVEs communicating with this virtual machine uses the old ARP
   entry.  If any VM in those NVEs need to talk to the new VM in the
   destination NVE, it uses the old ARP entry."
Repetition: these 2 sentences say the same. (The mistake is also repeated when these 2 sentences are repeated for IPv6).

s/Push the new mapping to hosts./Push the new mapping to communicating hosts./

The IPv4/IPv6 pairs of paras for "tunnel estabilshment" and "tunneling packets"
only differ in the words "IPv4"/"IPv6". So in each case a single para could be given for IP (irrespective of whether v4 or v6).

Thank you very much.

Linda Dunbar