Internet DRAFT - draft-liu-multicast-for-computing-storage
draft-liu-multicast-for-computing-storage
Network Working Group Y. Liu
Internet-Draft China Mobile
Intended status: Informational X. Geng
Expires: 11 January 2024 Huawei
10 July 2023
Multicast for Computing and Storage
draft-liu-multicast-for-computing-storage-00
Abstract
This document introduces the multicast use case for computing and
storage.
Requirements Language
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
"SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this
document are to be interpreted as described in RFC 2119 [RFC2119].
Status of This Memo
This Internet-Draft is submitted in full conformance with the
provisions of BCP 78 and BCP 79.
Internet-Drafts are working documents of the Internet Engineering
Task Force (IETF). Note that other groups may also distribute
working documents as Internet-Drafts. The list of current Internet-
Drafts is at https://datatracker.ietf.org/drafts/current/.
Internet-Drafts are draft documents valid for a maximum of six months
and may be updated, replaced, or obsoleted by other documents at any
time. It is inappropriate to use Internet-Drafts as reference
material or to cite them other than as "work in progress."
This Internet-Draft will expire on 11 January 2024.
Copyright Notice
Copyright (c) 2023 IETF Trust and the persons identified as the
document authors. All rights reserved.
This document is subject to BCP 78 and the IETF Trust's Legal
Provisions Relating to IETF Documents (https://trustee.ietf.org/
license-info) in effect on the date of publication of this document.
Please review these documents carefully, as they describe your rights
and restrictions with respect to this document. Code Components
Liu & Geng Expires 11 January 2024 [Page 1]
Internet-Draft draft-liu-multicast-for-computing-storag July 2023
extracted from this document must include Revised BSD License text as
described in Section 4.e of the Trust Legal Provisions and are
provided without warranty as described in the Revised BSD License.
Table of Contents
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . 2
2. Use in Multi-tenant Cloud . . . . . . . . . . . . . . . . . . 2
3. Typical Multicast Scenario in Computing . . . . . . . . . . . 3
3.1. AI Training . . . . . . . . . . . . . . . . . . . . . . . 3
3.2. HPC . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
4. Typical Multicast Scenario in Computing . . . . . . . . . . . 6
5. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 7
6. Security Considerations . . . . . . . . . . . . . . . . . . . 7
7. Acknowledgements . . . . . . . . . . . . . . . . . . . . . . 7
8. Normative References . . . . . . . . . . . . . . . . . . . . 7
Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . 8
1. Introduction
There are applications in data center with point-to-multipoint
communication patterns that would benefit from network multicast
service, without which, these applications, when migrating to public
clouds, will use server based packet replication techniques. This
leads to CPU load inflation and prevents tenants from sustaining high
throughputs and low latencies for multicast workloads.
In order to better show the requirements for computing and storage,
we list 3 typical potential multicast scenarios with P2MP services:
Multi-tenant Cloud, Computing and Storage.
The multicast requirements could be described with the following 3
aspects:
- Network Scale: number of switches, number of links, number of hosts
- Multicast Tree Size: number of intermeidate nodes; number of
receivers
- Multicast Service Number
2. Use in Multi-tenant Cloud
As illustrated in the following figure, a data-center may contain: a
network fabric configured in unicast-only mode, hosts running as
virtual machines (VMs) managed by tenants, central replicators (C-R)
for providing MSR6 packet delivery service among the hosts of a
tenant.
Liu & Geng Expires 11 January 2024 [Page 2]
Internet-Draft draft-liu-multicast-for-computing-storag July 2023
+----+ +----+ +----+
|C-R1| |C-R2| |C-Rn| ====>Central-Replicator
+-+--+ +-+--+ +-+--+
| | |
+-----+--------+--------+-----+
| Spine +Leaf +vSwitch Fabric |
| (Unicast Only) |
+--+--+--+--+--+--+--+--+--+--+
| | | | | | | |
h1 | h3 h4 | h6 | h8 ====>Tenant-1
| | |
h2 h5 h7 ====>Tenant-2
Take tenant-1 for example. The host h1 can send multicast flow using
MSR6 packets to C-R1, the MSR6 packets include one or more of the
destination hosts h3/h4/h6/h8 encoding in the MSR6 header. An MSR6
packet may be sent to C-R1 where it is replicated and sends to the
desired destination hosts. An MSR6 packet may be sent to C-R1 where
it is replicated and sends to the part of the destination hosts, and
another copy to C-R2 for replication and delivery to the left
destination hosts.
A Tenant may have a dedicated set of C-Rs for its own use, or a
Tenant may use a shared C-Rs for its replication requirement among
VMs.
3. Typical Multicast Scenario in Computing
3.1. AI Training
The following figure shows a typical RDMA AI training scenario.
PS(Parameter Server) Nodes
+-------+ +-------+
| CPU | | CPU |
| Server| | Server|
+-+-+-+-+ +-+-+-+-+
^ | | | | | | |
| +--|-|-|--------------+ | | |
| +----+ | +----------------------+ |
| | | +--------+ +-------+ | | V
Gradients | | | | | | Parameters
+---+-+-+ +---+-+-+ +-+---+-+
| GPU | | GPU | | GPU |
| Worker| | Worker| | Worker|
+-------+ +-------+ +-------+
Worker->PS: The gradient of each worker is pushed to PS node
Liu & Geng Expires 11 January 2024 [Page 3]
Internet-Draft draft-liu-multicast-for-computing-storag July 2023
PS->Worker: PS will pull the parameters back to all workers after
aggregation
In this process, the second stage is information distribution, with
the same data content. N connections are used to transmit unicast
separately. The bandwidth efficiency is 1/N, the larger the scale,
the lower the efficiency.
+---------------+
| Source |
| +---+ +---+ |
| |CPU| |GPU| |
| +-+-+ +-+-+ |
| | | |
| \ / |
| +-V---V-+ |
| | HCA | |
| +-------+ |
+--+-+-+-+-+-+--+
| | ... | |
+--V-V-----V-V--+
| Switch |
+-+-----------+-+
/ \
+-------------V-+ +-V------------+
| Destination | | Destination |
| +-------+ | | +-------+ |
| | HCA | | | | HCA | |
| +-V---V-+ | | +-V---V-+ |
| / \ | | / \ |
| | | | | | | |
| +-+-+ +-+-+ | | +-+-+ +-+-+ |
| |CPU| |GPU| | | |CPU| |GPU| |
| +---+ +---+ | | +---+ +---+ |
+---------------+ +---------------+
If the source only sends 1 copy to the network and the switches
replicate the packet to different distinations. The use of bandwidth
is more efficient and the training is faster.
The large-scale multicast requirement in this scenario is as the
following:
- Network Scale: 10-10k GPU
- Multicast Tree Size: 10-10k receivers
- Multicast Service Number: depends on the scenario
Liu & Geng Expires 11 January 2024 [Page 4]
Internet-Draft draft-liu-multicast-for-computing-storag July 2023
3.2. HPC
The following is an example of MPI in HPC scenario.
+-------------------------------------------+
| Dispatcher |
| Master |
+---------------------+---------------------+
|
+-----------------+
|
+---+----+ +--------+ +--------+
|+--V---+| |+------+| |+------+|
||Dispa-|| ||Dispa-|| ||Dispa-||
||Agent || ||Agent || ||Agent ||
|+---+--+| |+---+--+| |+---+--+|
| | | | | | | | |
|+---V--+| |+---V--+| |+---V--+|
|| MPI || || MPI || ... || MPI ||
||Proces|| ||Proces|| ||Proces||
|+---^--+| |+---^--+| |+---^--+|
| | | | | | | | |
|+---V--+| |+---V--+| |+---V--+|
|| RoCE |<-->| RoCE |<------------->| RoCE ||
|+------+| |+------+| |+------+|
+--------+ +--------+ +--------+
Stage 1: Dispatcher Master senses millions of cores and schedules
millions of Rank MPI jobs on demand. Dispatcher Master sends the
scheduling results to Dispatcher Agent
Stage 2: Dispatcher Agent starts Million Rank MPI on each node The
Dispatcher Agent that receives the message broadcast the message to
other Dispatcher Agents and do the initialization before starting the
MPI application
Stage 3: Dispatcher Agent broadcaast the message to start the MPI
application. MPI internal initialization Synchronize the RoCE
endpoint in allgather way after the MPI application is started
The last 2 stages could benefit from multicast and reduce task
completion time.
The large-scale multicast requirement in this scenario is as the
following:
- Network Scale: 1000 k CPU/GUP
Liu & Geng Expires 11 January 2024 [Page 5]
Internet-Draft draft-liu-multicast-for-computing-storag July 2023
- Multicast Tree Size: 10k~100k receivers
- Multicast Service Number: 1~100
4. Typical Multicast Scenario in Computing
Ceph is an open-source distributed software platform. It mainly
focuses on scale-out file system including storage distribution and
availability, which is widely used in storage.
Ceph Object Storage Daemons (OSDs) are reponsible for storing objects
on a local file system on behalf of Ceph clients. Also, Ceph OSDs
use the CPU, memory, and networking of Ceph cluster nodes for data
replication, erasure coding, recovery, monitoring and reporting
functions.
The following process request P2MP service.
- Application initiates "write" operation from a client to a server.
- Client finds the server to write in, and 3 copies are sent to 3
services.
+-------+ +-------+
|Client1| |Client2|
+---+---+ +---+---+
| |
+---------+--------+
|
+-------+-------+
| Switch |
+-------+-------+
|
+----------------+----------------+
| | |
+---+---+ +---+---+ +---+---+
| Server| | Server| | Server|
+-------+ +-------+ +-------+
The large-scale multicast requirement in this scenario is as the
following:
- Network Scale: 3k Server (1 Pod)
- Multicast Tree Size: 3 receivers
- Multicast Service Number: 10k
Liu & Geng Expires 11 January 2024 [Page 6]
Internet-Draft draft-liu-multicast-for-computing-storag July 2023
5. IANA Considerations
This document makes no request of IANA.
6. Security Considerations
TBD
7. Acknowledgements
TBD
8. Normative References
[I-D.cheng-spring-ipv6-msr-design-consideration]
Cheng, W., Mishra, G. S., Li, Z., Wang, A., Qin, Z., and
C. Fan, "Design Consideration of IPv6 Multicast Source
Routing (MSR6)", Work in Progress, Internet-Draft, draft-
cheng-spring-ipv6-msr-design-consideration-01, 25 October
2021, <https://datatracker.ietf.org/doc/html/draft-cheng-
spring-ipv6-msr-design-consideration-01>.
[I-D.ietf-avtcore-rtp-topologies-update]
Westerlund, M. and S. Wenger, "RTP Topologies", Work in
Progress, Internet-Draft, draft-ietf-avtcore-rtp-
topologies-update-10, 2 July 2015,
<https://datatracker.ietf.org/doc/html/draft-ietf-avtcore-
rtp-topologies-update-10>.
[RFC2119] Bradner, S., "Key words for use in RFCs to Indicate
Requirement Levels", BCP 14, RFC 2119,
DOI 10.17487/RFC2119, March 1997,
<https://www.rfc-editor.org/info/rfc2119>.
[RFC3493] Gilligan, R., Thomson, S., Bound, J., McCann, J., and W.
Stevens, "Basic Socket Interface Extensions for IPv6",
RFC 3493, DOI 10.17487/RFC3493, February 2003,
<https://www.rfc-editor.org/info/rfc3493>.
[RFC3542] Stevens, W., Thomas, M., Nordmark, E., and T. Jinmei,
"Advanced Sockets Application Program Interface (API) for
IPv6", RFC 3542, DOI 10.17487/RFC3542, May 2003,
<https://www.rfc-editor.org/info/rfc3542>.
[RFC8200] Deering, S. and R. Hinden, "Internet Protocol, Version 6
(IPv6) Specification", STD 86, RFC 8200,
DOI 10.17487/RFC8200, July 2017,
<https://www.rfc-editor.org/info/rfc8200>.
Liu & Geng Expires 11 January 2024 [Page 7]
Internet-Draft draft-liu-multicast-for-computing-storag July 2023
Authors' Addresses
Yisong Liu
China Mobile
Email: liuyisong@chinamobile.com
Xuesong Geng
Huawei
Email: gengxuesong@huawei.com
Liu & Geng Expires 11 January 2024 [Page 8]