托管磁盘突发Managed disk bursting

在 Azure 上,我们提供了在虚拟机和磁盘上提升磁盘存储 IOPS 和 MB/秒性能的功能(称为突发)。On Azure, we offer the ability to boost disk storage IOPS and MB/s performance referred to as bursting on both Virtual Machines and disks. 突发在许多情况下很有用,例如处理意外的磁盘流量或处理批处理作业。Bursting is useful in many scenarios, such as handling unexpected disk traffic or processing batch jobs. 可以有效地利用 VM 和磁盘级别的突发,以在 VM 和磁盘上实现良好的基线性能和突发性能。You can effectively leverage VM and disk level bursting to achieve great baseline and bursting performance on both your VM and disk. 这样,你就可以在 VM 和磁盘上都获得良好的基线性能和突发性能。This way you can achieve great baseline performance and bursting performance on both your vm and disk.

请注意,磁盘和 VM 上的突发相互独立。Please note that bursting on Disks and VMs are independent from one another. 如果有突发磁盘,则不需要使用突发 VM 来允许磁盘突发。If you have a bursting disk, you do not need a bursting VM to allow your disk to burst. 如果有突发 VM,则不需要使用突发磁盘来允许 VM 突发。If you have a bursting VM, you do not need a bursting disk to allow your VM to burst.

Azure 高级 SSD 提供以下各种突发模型:Azure premium SSDs offer the following models of bursting:

  • 基于额度的模型:仅在磁盘的额度桶中积满突发额度时,磁盘才会突发。A credit-based model, where the disk will burst only if it has burst credits accumulated in its credit bucket. 当磁盘突发时,此模型不会产生额外的费用。This model does not incur additional charges when the disk bursts. 基于额度的突发仅适用于不超过 512 GiB 的磁盘。Credit-based bursting is only available on disks 512 GiB and smaller.

此外,可以更改托管磁盘的性能层,当工作负载在突发中运行时,这是理想的选择。Additionally, the performance tier of managed disks can be changed, which could be ideal if your workload would otherwise be running in burst.

基于额度的突发Credit-based bursting 更改性能层Changing performance tier
方案Scenarios 适用于短期缩放(不超过 30 分钟)。Ideal for short-term scaling (30 minutes or less). 如果工作负载将继续在突发中运行,那么这是理想之选。Ideal if your workload would otherwise continually be running in burst.
CostCost 免费Free 每个性能层的成本是固定的,请参阅托管磁盘定价,了解详细信息。The cost of each performance tier is fixed, see Managed Disks pricing for details.
可用性Availability 仅适用于高级 SSD 512 GiB 和更小的磁盘。Only available for premium SSDs 512 GiB and smaller. 适用于所有高级 SSD 大小。Available to all premium SSD sizes.
支持Enablement 在符合条件的磁盘上默认启用。Enabled by default on eligible disks. 用户必须手动更改其层级。User must manually change their tier.

常见场景Common scenarios

以下需求场景可显著受益于突发:The following scenarios can benefit greatly from bursting:

  • 缩短启动时间 - 在使用突发的情况下,实例的启动速度将大大加快。Improve startup times - With bursting, your instance will startup at a significantly faster rate. 例如,启用高级层的 VM 的默认 OS 磁盘是 P4 磁盘,其预配性能最高可达 120 IOPS 和 25 MB/s。For example, the default OS disk for premium enabled VMs is the P4 disk, which is a provisioned performance of up to 120 IOPS and 25 MB/s. 在使用突发的情况下,P4 的性能最高可达 3500 IOPS 和 170 MB/s,可使启动速度提高达 6 倍。With bursting, the P4 can go up to 3500 IOPS and 170 MB/s allowing for startup to accelerate by up to 6X.
  • 处理批处理作业 - 一些应用程序工作负载本质上是周期性的。Handle batch jobs - Some application workloads are cyclical in nature. 它们在大多数时间需要基线性能,只在短时间内需要更高的性能。They require a baseline performance most of the time, and higher performance for short periods of time. 这种情况的一个示例是一个会计程序,该程序每日处理需要少量磁盘流量的事务。An example of this is an accounting program that processes daily transactions that require a small amount of disk traffic. 该程序会在月末完成需要较高磁盘流量的对帐报表。At the end of the month this program would complete reconciling reports that need a much higher amount of disk traffic.
  • 流量高峰 - Web 服务器及其应用程序随时都可能遇到流量激增。Traffic spikes - Web servers and their applications can experience traffic surges at any time. 如果 Web 服务器由使用突发的 VM 或磁盘提供支持,那么这些服务器就能更好地处理流量高峰。If your web server is backed by VMs or disks that use bursting, the servers would be better equipped to handle traffic spikes.

磁盘级突发Disk-level bursting

基于额度的突发Credit-based bursting

对于 Azure 中国世纪互联的所有区域中大小为 P20 的磁盘和更小的磁盘,我们提供了基于额度的突发。Credit-based bursting is available for disk sizes P20 and smaller in all regions in Azure China 21Vianet. 在受支持磁盘大小的所有新的和现有的部署上,默认已启用磁盘突发。By default, disk bursting is enabled on all new and existing deployments of supported disk sizes. VM 级突发只使用基于额度的突发。VM-level bursting only uses credit-based bursting.

虚拟机级突发Virtual machine-level bursting

VM 级突发只使用基于额度的模型进行突发,默认情况下会为支持此功能的所有 VM 启用。VM-level bursting only uses the credit-based model for bursting, it is enabled by default for all VMs that support it. 为 Azure 中国世纪互联的所有区域中以下受支持的大小启用了 VM 级突发:VM-level bursting is enabled in all regions in the Azure China 21Vianet on these supported sizes:

突发流Bursting flow

突发额度系统以相同的方式应用于 VM 级别和磁盘级别。The bursting credit system applies in the same manner at both the VM level and disk level. 你的资源(VM 或磁盘)的突发 Bucket 内的储备额度一开始是满的。Your resource, either a VM or disk, will start with fully stocked credits in its own burst bucket. 有了这些额度,你就能够以最大的突发速率进行最多 30 分钟的突发。These credits allow you to burst for up to 30 minutes at the maximum burst rate. 每当利用的资源的 IOPS 或 MB/秒低于资源的性能目标时,就会积累额度。You accumulate credits whenever the resource's IOPS or MB/s are being utilized below the resource's performance target. 如果你的资源已经积累了突发额度,并且你的工作负荷需要额外的性能,则你的资源可以使用这些额度来超出其性能限制,并提高其性能以满足工作负荷需求。If your resource has accrued bursting credits and your workload needs the extra performance, your resource can use those credits to go above its performance limits and increase its performance to meet the workload demands.

突发 Bucket 示意图。

你如何使用可用额度取决于你自己。How you spend your available credits is up to you. 你可以连续使用 30 分钟的突发额度,也可以在一天内分散地使用。You can use your 30 minutes of burst credits consecutively or sporadically throughout the day. 当你部署资源时,系统会为它们分配全部的额度。When resources are deployed they come with a full allocation of credits. 当它们耗尽时,只需要不到一天的时间进行补充。When those deplete, it takes less than a day to restock. 你可以自行决定如何使用额度,但突发 Bucket 不需要充满,资源便可突发。Credits can be spent at your discretion, the burst bucket does not need to be full in order for resources to burst. 突发累积因资源而异,因为它基于资源在未达到性能目标的情况下运行时未使用的 IOPS 和 MB/秒。Burst accumulation varies depending on each resource, since it is based on unused IOPS and MB/s below their performance targets. 与基线性能较低的资源相比,基线性能较高的资源可以更快地积累其突发额度。Higher baseline performance resources can accrue their bursting credits faster than lower baseline performing resources. 例如,一个闲置的 P1 磁盘每秒将积累 120 IOPS,而一个闲置的 P20 磁盘每秒将积累 2,300 IOPS。For example, a P1 disk idling will accrue 120 IOPS per second, whereas an idling P20 disk would accrue 2,300 IOPS per second.

突发状态Bursting states

启用了突发功能时,资源可能处于以下三种状态之一:There are three states your resource can be in with bursting enabled:

  • 正在积累 - 资源正在使用的 IO 流量低于性能目标。Accruing - The resource's IO traffic is using less than the performance target. 为 IOPS 和 MB/s 累积额度是彼此分开执行的。Accumulating bursting credits for IOPS and MB/s are done separate from one another. 你的资源可能会积累 IOPS 额度并支出 MB/s 额度,也可能会支出 IOPS 额度并积累 MB/s 额度。Your resource can be accruing IOPS credits and spending MB/s credits or vice versa.
  • 正在突发 - 资源正在使用的流量高于性能目标。Bursting - The resource's traffic is using more than the performance target. 突发流量将独立消耗 IOPS 或带宽的额度。The burst traffic will independently consume credits from IOPS or bandwidth.
  • 恒定 - 资源的流量与性能目标完全相同。Constant - The resource's traffic is exactly at the performance target.

突发示例Bursting examples

以下示例显示了在使用各种 VM 和磁盘组合时突发的工作情况。The following examples show how bursting works with various VM and disk combinations. 为了使示例易于理解,我们将重点放在 MB/s 上,但相同的逻辑也独立适用于 IOPS。To make the examples easy to follow, we will focus on MB/s, but the same logic is applied independently to IOPS.

具有可突发磁盘的非可突发虚拟机Non-burstable virtual machine with burstable disks

VM 和磁盘组合:VM and disk combination:

  • Standard_D8s_v3Standard_D8s_v3

    • 未缓存的 MB/s:192Uncached MB/s: 192
  • P4 OS 磁盘P4 OS Disk

    • 预配的 MB/s:25Provisioned MB/s: 25
    • 最大突发 MB/s:170Max burst MB/s: 170
  • 2 个 P10 数据磁盘2 P10 Data Disks

    • 预配的 MB/s:100Provisioned MB/s: 100
    • 最大突发 MB/s:170Max burst MB/s: 170

当 VM 启动时,它从 OS 磁盘中检索数据。When the VM boots up it retrieves data from the OS disk. 由于 OS 磁盘是正在启动的 VM 的一部分,因此 OS 磁盘的突发额度将是充分的。Since the OS disk is part of a VM that is booting, the OS disk will be full of bursting credits. 这些额度使 OS 磁盘以 170 MB/s 的速度突发启动。These credits will allow the OS disk burst its startup at 170 MB/s second.

VM 将 192 MB/s 的吞吐量请求发送到 OS 磁盘,OS 磁盘以 170 MB/s 数据进行响应。

启动完成后,应用程序将在 VM 上运行,并且具有非关键工作负荷。After the boot up is complete, an application is then run on the VM and has a non-critical workload. 此工作负荷需要 15 MB/S(在所有磁盘上均匀分布)。This workload requires 15 MB/S that gets spread evenly across all the disks.

应用程序向 VM 发送 15 MB/s 的吞吐量请求,VM 接收请求并向其每个磁盘发送 5 MB/s 的请求,每个磁盘返回 5 MB/s,VM 将 15 MB/s 返回到应用程序。

然后,应用程序需要处理一个批处理作业,该作业需要 192 MB/s。Then the application needs to process a batched job that requires 192 MB/s. 2 MB/s 由 OS 磁盘使用,其余部分在数据磁盘之间平均划分。2 MB/s are used by the OS disk and the rest are evenly split between the data disks.

应用程序将 192 MB/s 的吞吐量请求发送到 VM,VM 接收请求,并将其大部分请求发送到数据磁盘(每个 95 MB/s)并将 2 MB/s 发送到 OS 磁盘,数据磁盘突发以满足需求,并且所有磁盘都将请求的吞吐量返回给 VM,后者将其返回给应用程序。

后续步骤Next steps

若要了解如何深入了解突发资源,请参阅磁盘突发指标To learn how to gain insight into your bursting resources, see Disk bursting metrics.