Editor’s note: This part 6 of 6 of a study shows in detail attempts to improve the loading budget, layer 7. Read the rest below.
Links table
6 Conclusion
We suggest Laconic, which explores the L7 emptying capabilities to download the budget on Smartnics and address its associated challenges. In Laconic, we use a lightweight network staple to avoid the costs of a heavy network staple, which is rich in features. To reduce the cost of synchronization, we design lightweight synchronization mechanisms that allow the synchronization and expansion high. For more reduce the burden on the general core on the smart, we take out the package processing by making the effective benefit from the shedding engines available on Smartnics. This paper does not raise any moral issues.
Reference
[1] Broadcom Ps1100R.
[2] Elb opposite ALB opposite NLB: Choose the best AWS download for your needs. https://iamondemand.com/blog/elb-vs-alb-vs-nlb-choosesthe-best-aws-low- billiser-for-yur-nes/.
[3] Photography can solve the problem of the Cloud Hotel Hotel in California. https://blocksandfiles.com/2021/07/19/fungible-colve-thepublic-cloud-trillion-dollar-paradox-hotel-california-plarm/.
[4] How to reduce the AWS Elb costs by 90 % using the application download budgets. . 78d4E980624B.
[5] Knowing the amount that must be spent on AWS. https://logz.io/blog/cost- Management-elb-aws-low-belancer/.
[6] Reducing the TCO data center with server emptying strategies. Dataceernmynamics.com/en/opinions
[7] What is the application of the application. https://docs.aws.amazon.com/elasticloadbalancing.
[8] Akhan Akulut and Harry C. Pyros. Small service design performance analysis. IEEE Internet Computing, 23 (6): 19-27, 2019.
[9] Mohamed Elizadeh, Tom Edsal, Sarang Darmaboricar, Ramanan Vidyanathan, Kevin Chu, Andy in Vinhout, Vinna Lam, Francis Matos, Rong Ban, Navindra Yadaf, and George Vargiz. Christ: Distinguished Data Data Download Budget. In the facts of the 2014 ACM conference on Sigcomm, Sigcomm ’14, page 503-514, New York, New York, USA, 2014. Computing machines Association.
[10] Broadcom. Smingray and IC smart transformers. [EB/OL]. https: //www.broudcom.com/products/ethernet-connectivity/netWorkadapters/smartnic reached on October 25, 2020.
[11] Qizhe Cai, Midhul Vupalapati, Jaehyun Hwang, Christos Kozyrakis, and Rachit Agarwal. Towards the time of the tail transmission and Terabit ethernet: emptying the host’s host. In the facts of the ACM Sigcomm 2022 conference, pages 767-779, 2022.
[12] DPDK. [rfc] Ethdev: Support the hairpin waiting menu. [EB/OL]. [email protected]/ It was reached on October 20, 2021.
[13] Dropbox. How did we migrate from NGINX to the envoy. [EB/OL]. Infrastructure/How Robox-Robox-Mennx was accessed to Evevoy on April 19, 2022.
[14] Daniel e Eyssinbod, Cheng Yi, Carlo Contefali, Cody Smith, Roman Conunov, Eric Man Hilcher, Ardas Celinjojo, Ben Cheney, Mino Shang, and Jinnah Dylan Hosen. Maglev: Fast and reliable program balance. At the 13th {usenix} symposium on the design and implementation of the existing systems ({NSDI} 16), pages 523-535, 2016.
[15] Inc F5. NGINX High Performal Load Balance. [EB/OL]. Nginx.com/ It was reached on October 25, 2020.
[16] Facebook. Catheran. [EB/OL]. Catheran was reached on October 25, 2020.
[17] Roy Fielding, Jim Gytez, Jeffrey Mogol, Henrik Freuzic, Larry Massenter, Paul Leich, and Tim Burners Lee. RFC2616: Hetter Transmission Protocol – HTTP/1.1, 1999.
[18] The original computing cloud. The agency’s envoy. [EB/OL]. Envoyproxy.io/ reached on October 25, 2020.
[19] Soda Gorbani, Godfrey, Yashar Ganjali, and Amin Veruwozasian. The exact load balance in data centers with drilling. In the facts of the fourteenth ACM workshop on hot topics in networks, pages from 1 to 7, 2015.
[20] Soda Gorbani, Zeepin Yang, P. Godfrey, Yashar Gangali, and Amin Firuzahian. Drilling: The exact load balance of the low data center networks. In the facts of the ACM Special Interest Group Conference on Data Communication, Pages 225-238, 2017.
[21] Massimo Jerondi, Marco Chisa, and Tom Barbit. Follow the high -speed connection in modern servers. In 2021 IEEE 22nd International Conference on Switching and Guidance Higher Performance (HPSR), pages 1-8. IEEE, 2021.
[22] Glasser will. WRK-A HTTP Benchmarking Tool, 2018.
[23] Google. Google cloud end points are now generally available: Fast and developed API gate. [EB/OL]. GCP/Google-CLOUDPONTS-NOW-GA-A-Fast-SCAL-Gateway that was reached on October 25, 2020.
[24] Haberoxi. Hiproxy reliable download TCP/HTTP reliable and high performance. [EB/OL]. HPROXY.org/ It was reached on October 25, 2020.
[25] Eunyoung Jeong, Shinae Wood, Muhammad Jamshed, Hawon Jeong, Sunghwan Ihm, Dongsu Han and KyoungSoo Park. MTCP: Stack {TCP} at the highly developmental user level of multiple systems. At the eleventh symposium {Usenix} on the design and implementation of network systems ({nsdi} 14), pages 489-502, 2014.
[26] Georgios P Katsikas, Tom Barbette, Marco Chiesa, Dejan Kostić and Gerald Q Maguire. What you need to know about the (smart) network cards. At the International Conference for the Measurement of negative and active network, pages 319-336. Springer, 2021.
[27] Naga Kata, Mokish Hera, Zhangon Kim, Enerod Sivaraman, and Jennifer Rixford. Hola: The downloadable download budget using programmable data planes. In the symposium’s facts about SDN Research, pages 1-12, 2016.
[28] Antoine Kaufman, Tim Store, Simon Peter, Navin Kerr Sharma, Arfind Krishnamori, and Thomas Anderson. TAS: TCP accelerates as an operating system service. In the facts of the Fourteenth Eurosys Conference 2019, pages from 1 to 16, 2019.
[29] Duckwoo Kim, Seungon Lee, and KyoungSoo Park. A case for private communications that accelerate smart. In the Fourth Asia Pacific Workshop on Networks, Pages 30-35, 2020.
[30] Taehyun Kim, Deondre Martin Ng, Junzhi Gong, Youngjin Kwon, Minlan Yu and Kyoungso Park. RearkeGinging TCP Cottact to deliver I/Ooffloaded content. At the nineteenth Usenix symposium on the design and implementation of the network connected to the network, NSDI 2022. Usenix, 2023.
[31] Bojie Li, Tianyi Cui, Zibo Wang, Wei Bai and Lintao Zhang. Socksdirect: The database sockets can be fast and compatible. At the ACM Sigcomm Conference (Sigcomm), August 2019.
[32] Bojie Li, Kun Tan, Layong (LURO) LUO, Yanqing Peng, Renqian Luo, Niningyi Xu, Yongqiang Xiong, Peng Cheng, and Opnong Chen. Clicknp: Treating a very flexible and high -performance network with reinformed devices. In the facts of the 2016 ACM Sigcomm Conference, Sigcomm ’16, page 1-14, New York, New York, USA, 2016. Computing machines Association.
[33] Xiaozhou Li, David G Andersen, Michael Kaminsky and Michael J Freedman. The algorithm improvements to seduce the rollers. In the facts of the ninth European Conference on Computer Systems, pages 1 to 14, 2014.
[34] Gianchen Leo, Carlos Malzan, Craig Olmeer, and Matthew Leon Carey. Performance characteristics of Bluefield-2 smart. Arxiv Preprint Arxiv: 2105.06619, 2021.
[35] Ming Liu, Tianyi Cui, Henry Schuh, Arvind Krishnamurthy, Simon Peter, and Karan Gupta. Emptying applications distributed on Smartnics using IPipe. In the facts of the ACM private interest group on data connection, pages 318-333. 2019.
[36] Michael Marti, Mark de Crevaf, Jacob Adrianz, Christopher Alfield, Sean Power, Carlo Conthavali, Michael Dalton, Nandita Dukibati, William C Evans, Steve Griel, and others. Snap: Microkernel approach to host networks. In the facts of the ACM 27 seminar on the principles of operating systems, pages 399-413, 2019.
[37] marvel. Octeon Tx2 Liquidio III Smartnic. [EB/OL]. Marvell.com/products/infracture-Proesssors
[38] Roy Miao, Huni Zing, Changon Kim, Jeongkon Lee, and Minlan Yu. Silkroad: Make a government layer 4 fast and cheap budget using ASIC. In the facts of the ACM Special Interest Group Conference on Data Communication, Pages 15-28, 2017.
[39] Younggyoun Moon, Seungon Lee, Muhammad Asim Jamshed, and Kyoungso Park. Accesscp: Speeding network applications with TCP discharge. At the seventeenth Usenix symposium on the design and implementation of the network connected to the network (NSDI 20), pages 77-92, Santa Clara, California, February 2020.
[40] Netronome. Agilio Cx Smartnics. [EB/OL]. Netronome. com/Products/Agilio-CX/reached on October 25, 2020.
[41] Nafidia. Bluefield Smartnic Eternet. [EB/OL]. Milanox. Com/Products/Bluefield-Smartnic-Etternet reached on October 25, 2020.
[42] Vladimir Olino, Alexandro Akturi, Andrei Phoenico, and Coston Resio. No database loader with beamer. In 15th {Usenix} A symposium on the design and implementation of the existing systems ({nsdi} 18), pages 125-139, 2018.
[43] PARVEEN PATEL, Deepak Bansal, Lihua Yuan, Ashwin Mresthy, Albert Greenberg, David A Maltz, Randy Kern, Hemant Kumar, Marios Zikos, Hongyu Wu, Et Al. Ananta: The cloud is the budget loading scale. ACM Sigcomm Computer Communication Review, 43 (4): 207-218, 2013.
[44] Rajath SHASHIDHARA, Tim Stamler, Antoine Kaufmann and Simon Peter. Flexoe: Flexible TCP emptying with careful parallel. At the nineteenth Usenix symposium on the design and implementation of the network systems (NSDI 22), pages 87-102, Renton, Washington, April 2022. Usenix Association.
[45] Chaoliang Zeng, Layong Luo, Teen Zhang, Zilong Wang, Luyang Li, Wenchen Han, Nan Chen, LEBING WAN, Lichao Liu, ZHIPING Ding, XionGfei Geng, Tao Feng, Feng Ning, Kai Chen, and Chuanxion Uo. Tiara: Speaking and effective budget for the loading budget with a government class. At the nineteenth Usenix symposium on the design and implementation of the network systems (NSDI 22), pages 1345-1358, Renton, Washington, April 2022. Usenix Association.
[46] Hong Chang, Junixoy Chang, Wei Bay, Kay Chen, and Mosharaf Chaudhry. Balancing a flexible data center in the wild. In the facts of the ACM Special Interest Group Conference on Data Communication, Pages 253-266, 2017.
[47] Nawa Zilberman. Technical perspective: HXDP: Emptying light and effective packages. Etisalat, 65 (8): 91-91, 2022.
extension
A.1 Laconic Productivity using all cores
Figure 13 shows Laconic productivity on Bluefield-2 with seven arm [1]. Bluefield-2 is equipped with 2 x 100 GB ports per second. We enable the perpetrators together to increase the emphasis on the treatment of the arm nuclei. For the NGINX foundation line that works on Commodity X86 servers, we use 2 x 100 GB ports per second from the Connectx-5 NIC to make the correlation speed the same. However, although there are 2 x 100 GB ports per second on the Connectx-5, due to the PCIE 3.0 lines of TestBed servers, only about 120 GB can achieve only. As shown in Figure 13, Laconic can achieve with seven nic nuclei only font rate[2] For large messages sizes. For small messages sizes, Laconic makes a similar performance with NGINX operation on X86. Bluefield-2 has a WIMPY arm nuclei, and NGINX works on ARM badly, as expected. There is a clear increase in the productivity of the messages that exceed 1 meter because Laconic separates the package in the hardware engine of the response messages greater than 1m. Unlike the first few beams, all the remaining beams in this flow are treated only by the process of processing devices packages. This can reduce the problem of banning head caused by a limited treatment force and also reduces the CPU resources for small or future responses. In the subsequent 4.4 section, we display experiences with different discharge sills.
Figure 14 shows the productivity of Laconic on Nicidio3 50g NIC with 24 curtain cores. Since Liqudio3 can support only 50 GB connection, we also discourage X86 foundation at 50 GB per second for a fair comparison of all experiences that work on Liquidio3. NGINX on the smart cores (LIO3 NGINX “) has low productivity. The productive plateaux is about 5 GB per second, even with 24 nuclei. Laconic can easily reach the NIC full reinforcement range with a 10K response. However, NGINX works on the X86 server needs 32 fat cores to achieve a similar performance.
[1] Bluefield-2 has eight arm core available in total. We keep one nucleus for the necessary background tasks.
[2] Since the same port on Smartnic receives data from each of the customer requests and also receives responses from the back interface servers, the theoretical limit in favor of BoodPut of the pregnancy balance is less than the 2x100GBPS line.
