Hi there,
in general terms, SAS indicates that all layers of a SAS environment must stay as close as possible, preferably in the same datacenter. Specifically to minimize network latency while increasing network throughput data transfers. Of course connections can be done towards outside, or from outside. But then you need to deal with all the aspects related to latency: timeouts, slowness, performance degradation, loss of packages and noise, etc. As such, you really want low latency numbers. As close as possible as being part of the same datacenter. Meaning, preferably within single digit milliseconds, 2 digits max (I'd say no much more than 20 ms). A (more complete) answer from AI, which should be taken with salt and pepper, but overall I can agree with it:
For a SAS environment with external connections, good latency means low single-digit milliseconds (ms) for interactive tasks like Enterprise Guide, while high throughput (hundreds of MB/s) is crucial for large data moves; slow network links (VPNs, high-latency internet) can severely degrade performance, emphasizing the need for fast, direct connections, ideally <20ms for cloud/remote, and very low internal SAS Grid latency for coordination, with <100ms overall being desirable for responsive user experience.
Key Latency & Throughput Factors
User Experience (SAS EG/VA):
<20ms: Excellent for fast, responsive work over LAN/direct cloud links.
20-50ms (Cable/Fast Internet): Generally acceptable but noticeable lag for some actions.
50-100ms+ (DSL/VPNs): Can cause significant delays (5-10s) due to aggregated small requests, frustrating users.
SAS Grid/Viya (Internal Nodes):
Very Low Latency: Essential for node-to-node coordination (e.g., 1-5ms).
High Throughput: Critical for loading data into memory (e.g., 100+ MB/s per core).
Cross-Prem/Cloud Data Movement:
Network Speed (Bandwidth): High latency can severely limit throughput (e.g., only 500KB/s to 20MB/s over some cloud links).
Physical Distance: Major contributor to latency; keep SAS and data close.
What's "Good"?
Aim for Sub-20ms: When connecting from outside (e.g., laptops, remote offices), aim for <20ms latency to the SAS environment (e.g., via dedicated circuits, ExpressRoute, fast VPNs) for a snappy feel.
Measure Throughput: Ensure you have hundreds of MB/s available for data loads, not just low latency.
Isolate Workloads: Use dedicated network paths (VNETs, subnets) for SAS nodes to avoid contention and latency spikes from other traffic.
In summary, think beyond just latency: low latency provides responsiveness, but adequate network throughput is vital for moving large datasets efficiently. For external users, <20ms latency with high bandwidth (100s of MB/s) is the goal for a good experience
... View more
