none
High server latency - Up to 2000ms (not talking E2E)

    Question

  • Hi,

    While profiling my app under development, I realized that some of my storage queries had high Time To First Byte. So I downloaded the storage metrics to discover that AverageServerLatency could be very high (above 2000ms). In the logs I could find several resquest with server latency above 2000ms.

    A few info:

    • I'm the only user, I'm based in France
    • The location of the storage account is West Europe, North Europe
    • The account is a standard general purpose storage account
    • I'm only using block blobs (images)
    • I query the store through the MS storage SDK which uses SAS

    Happy to provide more info that could help understand why I encounter such high latencies.


    Sunday, September 18, 2016 6:22 PM

All replies

  • Hi,

    In the case of high AverageServerLatency for blob download requests, you should use the Storage Logging logs to see if there are repeated requests for the same blob (or set of blobs). For blob upload requests, you should investigate what block size the client is using (for example, blocks less than 64K in size can result in overheads unless the reads are also in less than 64K chunks), and if multiple clients are uploading blocks to the same blob in parallel. You should also check the per-minute metrics for spikes in the number of requests that result in exceeding the per second scalability targets.

    For more information, please refer this article:
    https://azure.microsoft.com/en-in/documentation/articles/storage-monitoring-diagnosing-troubleshooting/#metrics-show-high-AverageServerLatency

    Regards,
    Azam Khan

    Monday, September 19, 2016 3:57 PM
    Moderator
  • Thank you but as you can imagine I already went through the doc :-).

    I observe high latency on get queries. The suggestion from the doc to use CDN is not relevant. I'm not supposed to get server latency of 3sec. The order of magnitude should be ms.

    Monday, September 19, 2016 4:02 PM