谷歌云 Z4M 存储优化机型正式商用：本地固态盘最高 168,000 GiB、400 Gbps 云 RDMA 网络，随机读最高 1560 万 IOPS
谷歌云 11 月 7 日宣布，新一代存储优化机型系列 Z4M 在 Google Compute Engine 正式商用（GA），同时提供虚拟机与裸金属两种形态。谷歌称，Z4M 为超级智能与机器学习的训练、推理负载而建，覆盖检查点保存、分布式键值缓存与并行文件系统等需要海量数据吞吐的场景，也服务于业务关键型数据库。
配置方面，Z4M 采用第五代英特尔至强可扩展处理器，配合谷歌 Titanium 卸载与存储技术，单机本地固态盘（LSSD）容量最高 168,000 GiB，云 RDMA 网络带宽最高 400 Gbps；谷歌称 RDMA 相比 TCP 网络可把数据访问延迟最多降低 60%。Z4M 虚拟机与裸金属实例最高提供 192 个虚拟 CPU 与 1,536 GiB 内存，每虚拟 CPU 对应 875 GiB 本地固态盘容量；随机读性能最高 1560 万 IOPS，顺序读吞吐最高 75,600 MiB/s，谷歌称两项指标较上一代 Z3 分别最多提升 70% 与 110%。
维护与存储配套方面，本地固态盘不超过 42,000 GiB 的 Z4M 虚拟机可在维护事件中实时迁移，更大配置与裸金属实例须终止并重启、数据保留；Z4M 同时支持谷歌云块存储 Hyperdisk，单实例网络附加存储容量最高 512 TiB。合作伙伴表态方面，Nutanix 产品管理副总裁 Saveen Pakala 表示将把 Nutanix Cloud Clusters（NC2）带到 Z4M 裸金属实例；Silk 首席研发官 Adik Sokolovski 表示其测试中云 RDMA 在 Z4M 上的延迟最多低 3.7 倍、CPU 占用最多低 6 倍；WEKA 首席产品官 Ajay Singh 表示其测试显示 Z4M 顺序读吞吐约为上一代两倍、IOPS 最多高 79%。谷歌称 Z4M 即日起在全球部分区域可用，具体区域需向销售团队确认，公告未公布定价。
Google Cloud Z4M Storage-Optimized Machines Reach General Availability: Up to 168,000 GiB of Local SSD, 400 Gbps Cloud RDMA and 15.6 Million Random-Read IOPS
Google Cloud announced on November 7 that its next-generation storage-optimized Z4M machine series is now generally available in Google Compute Engine, offered as both virtual machines and bare-metal instances. Google said Z4M was built for super intelligence and machine learning training and inference workloads such as checkpointing, distributed key-value caches and parallel file systems, as well as business-critical databases.
Powered by fifth-generation Intel Xeon Scalable processors and Google's Titanium offload and storage technology, Z4M provides up to 168,000 GiB of local SSD storage coupled with up to 400 Gbps of Cloud RDMA networking, which Google said reduces data-access latency by up to 60% compared with TCP networking. Z4M VMs and bare-metal instances offer up to 192 vCPUs and up to 1,536 GiB of memory, with 875 GiB of local SSD per vCPU. Google said Z4M delivers up to 15.6 million random-read IOPS and up to 75,600 MiB/s of sequential-read throughput, improving on the previous Z3 series by up to 70% and 110% respectively.
For maintenance, Z4M VMs with 42,000 GiB or less of local SSD can be live-migrated, while larger VMs and bare-metal instances must be terminated and restarted with data preserved. Z4M also supports Hyperdisk block storage, with up to 512 TiB of network-attached capacity per instance. Among partner comments in the announcement, Nutanix vice president of product management Saveen Pakala said Nutanix Cloud Clusters (NC2) is coming to Z4M bare-metal instances; Silk chief R&D officer Adik Sokolovski said testing showed up to 3.7 times lower latency and up to 6 times lower CPU usage with Cloud RDMA over TCP; and WEKA chief product officer Ajay Singh said testing found roughly twice the read throughput and up to 79% more IOPS than the Z3 generation. Google said Z4M is available today in select regions worldwide, with regional availability confirmed through its sales team; the announcement listed no pricing.
来源：谷歌云
发布：GSM·美国（发布链接）
图片来源：图：Google 标志。Wikimedia Commons，公共领域（PD-textlogo）。 / Photo: Google logo. Wikimedia Commons, public domain (PD-textlogo).
GSM Global · 本篇英文版 English edition of this story
