Alle paar Monate kommt eine neue Liste der Unterauftragnehmer und diesmal zu Beginn März erscheint die aktuelle Liste vom 28. Februar 2025. Es gibt keine großen Änderungen, wie dass nun die Ansprechpartner verzeichnet sind, sondern nur eine zusätzlichen Provider aus den USA und ein Update der Adresse.
Änderungen
| Amplitude, Inc. (USA) | Als zusätzlicher Anbieter hinzugefügt | Azure PaaS, Azure Platform Services, Azure AI, Azure Developer Tooling and SDKs |
Customer experience (CX) analytic |
| Blueprint Technologies (USA) | Adresse aktualisiert | 2600 116th Avenue NE, First Floor, Bellevue WA 98004 US |
stellt Personal |
Download
https://servicetrust.microsoft.com/DocumentPage/f2cc90fe-471f-4f07-a9ae-af71ebce8486
SLA März 2025
| Additions/Updates | Deletions |
| Exchange Online Protection | Azure Applied AI Services |
| Azure AI Services | Azure Machine Learning Studio (classic) |
| Azure OpenAI Service | Appendix B – Service Level Commitment for Uptime |
| Traffic Manager Service | |
| Azure VMware Solution |
Azure AI Services
Additional Definitions:
“Total Transaction Attempts” is the total number of authenticated API requests by Customer during an Applicable Period for a given Azure AI Service API. Total Transaction Attempts do not include API requests that return an Error Code that are continuously repeated within a five-minute window after the first Error Code is received.
“Failed Transactions” is the set of all requests to the Azure AI Service API within Total Transaction Attempts that return an Error Code. Failed Transaction Attempts do not include API requests that return an Error Code that are continuously repeated within a five-minute window after the first Error Code is received.
“Uptime Percentage” for each API Service is calculated as Total Transaction Attempts less Failed Transactions divided by Total Transaction Attempts in an Applicable Period for a given API subscription.
Uptime Percentage is represented by the following formula:
Service Credit:
The following Service Levels and Service Credits are applicable to Azure AI Services APIs (except Azure OpenAI):
| Uptime Percentage | Service Credit |
| < 99.9% | 10% |
| < 99% | 25% |
Service Level Exceptions: A separate SLA applies to Azure OpenAI Service.
Azure OpenAI Service
Uptime Calculation and Service Levels for Azure OpenAI Service
Additional Definitions:
“Azure OpenAI Resource” refers to an Azure resource of type Azure OpenAI created in an Azure region in a Microsoft Azure subscription.
“Deployment” is a model endpoint deployed in an Azure OpenAI Resource.
“Request” is an API call to a model endpoint in an Azure OpenAI Resource.
“Average Error Rate” for an Applicable Period is the sum of Error Rates for each minute in the Applicable Period divided by the total number of minutes in the Applicable Period.
“Error Rate” is the total number of Requests that return an Error Code divided by the total number of Requests during a minute. If the total number of Requests in a given minute interval is zero, the error rate for that interval is 0%.
„Uptime Percentage“ is represented by the following formula
Service Credit:
| Uptime Percentage | Service Credit |
| < 99.9% | 10% |
| < 99% | 25% |
Latency Calculation and Service Levels for Azure OpenAI Provisioned Throughput Managed Deployments
Additional Definitions
“Azure OpenAI Deployment” is an Azure OpenAI model deployment for an applicable model within an Azure OpenAI resource.
“Azure OpenAI Provisioned Managed Deployment” is an Azure OpenAI model deployment configured to use the provisioned managed capability offering reserved model processing capacity with an SLA latency target value as specified in Azure OpenAI Provisioned Managed Deployment documentation.
“Maximum Request Size” is the maximum number of input and output tokens that can be included in a given request for a specified Azure OpenAI model, as specified in Azure OpenAI Models documentation.
“N” is the number of Successful Requests made to an Azure OpenAI Provisioned Throughput Managed deployment with a maximum request size that is less than or equal to the limit provided for the specified Azure OpenAI model in a given minute.
“Tokens Per Second” is a latency measurement that enumerates the generation speed for a given Azure OpenAI model response. The total tokens generated is divided by the time to generate the tokens. Generated tokens are counted as all tokens generated after the first token.
“SLA Latency Target Value” is the latency target in Tokens Per Second for a given model as specified in Azure OpenAI Provisioned Managed Deployment documentation.
“S” are the response times in N that meet the following criteria and are determined as follows:
- Tokens Per Second for calls generating more tokens than the SLA Latency Target Value OR taking more than 1 second to generate all response
- SLA Latency Target Value for calls generating less than or equal to the SLA Latency Target Value AND taking less than 1 second to generate all response tokens
- 0 for all calls generating 1 or fewer tokens
“Average Tokens Per Second” is the average of the Tokens Per Second for all S in a given minute.
“Excessive Latency Minutes”: is the total number of one-minute intervals during which the Average Tokens Per Second is less than the SLA Latency Target Value. If N is zero, the Excessive Latency Minutes for that interval is 0.
“Average Excessive Latency Rate” for an Applicable Period is the sum of Excessive Latency Minutes divided by the total number of minutes in the Applicable Period.
“Average Attainment Percentage” for a given Azure OpenAI Provisioned Throughput Managed Deployment is calculated by subtracting from 100% the Average Excessive Latency Rate for a given Azure OpenAI resource in an Applicable Period.
Time Between Token Latency Attainment Percentage is represented by the following formula:
Service Credit for Azure OpenAI Provisioned Throughput Managed Deployments
| Average Attainment Percentage | Service Credit |
| < 99% | 10% |
Exchange Online Protection
Downtime: Any period of time Microsoft 365 is not able to accept or send email messages due to a problem within the Microsoft 365 boundary. There is no Scheduled Downtime for this service.
Exchange Online Protection Availability is divided into two categories:
- Permanent rejections due to Microsoft 365 issues
- Transient rejections due to Microsoft 365 issues
Uptime Percentage
- Permanent rejections due to Microsoft 365 issues.
The Uptime Percentage is calculated using the following formula:
where Downtime is duration of the issue (in minutes) multiplied by the fraction of the service that is not available summed over the calendar month.
- Transient rejections due to Microsoft 365 issues
Uptime Percentage
Impacted Messages are defined as messages delayed more than (10) minutes during the issue summed over the calendar month.
Service Credit:
| Uptime Percentage | Service Credit |
| < 99.9% | 25% |
| < 99% | 50% |
| < 95% | 100% |
Exchange Online Protection applies only to legitimate business email delivered to or sent by valid Microsoft 365 licensed email accounts. This does not apply to:
- Customer induced mail storms
- Bulk email (customer mailers, newsletters, etc.)
- Delivery of email to an archive
- Denial of Service attacks (DoS)
- Microsoft 365 Tenant misconfiguration
- Email delays where the failure mode is within the customers on-premises boundary and/or third-party service providers.
- Network latency between Microsoft 365 and end user email clients
- Mail that is throttled by Microsoft 365 to protect the health of the service or because a tenant has exceeded defined send and/or receive messaging limits.
- Messages de-prioritized by Microsoft 365 to protect the overall health of the service (For example: large messages, messages sent to a large number of recipients or distribution lists that contains substantial number of recipients)
- Messages delayed due to expensive customer configured rules and policies