Slow send message to SQS using Lambda with low memory - performance

Question

We have created some sample lambdas to test SQS insertion performance.
At first we used 128MB lambdas and got 40ms average for insertions of 7KBytes packages.
When we upgraded the lambda to 256MB we got 20ms averages and when we upgraded to 512MB we even got 11ms average time for insertion.
We wanted to know why the bigger memory/cpu of the lambda gets better insertion speeds on SQS, because we thought sqs insertion should not be related to memory or CPU capacity. In summary: Is there any operation inside the sqs client send operation that requires good CPU/Memory to perform better?
Thanks!
Here is our test code:
public class Function
{
public async Task<string> FunctionHandler(ILambdaContext context)
{
var client = new AmazonSQSClient();
var request = new SendMessageRequest
{
MessageAttributes = new Dictionary<string, MessageAttributeValue>(),
MessageBody = "7kbyte String here...",
QueueUrl = "https://our_account/sqs-test-queue "
};
//Just desconsidering the first 2 sent messages, to desconsider the initial connection overhead
var elapsed00 = await SendMessageResponse(client, request, -2);
var elapsed0 = await SendMessageResponse(client, request, -1);
double totalTime = 0;
const int totalIterations = 1000;
for (int j = 0; j < totalIterations; j++)
{
var elapsed = await SendMessageResponse(client, request, j);
totalTime += elapsed;
}
Console.WriteLine("$$$$$AVERAGE:" + ((double)(totalTime / totalIterations)));
return "End test";
}
private async Task<long> SendMessageResponse(AmazonSQSClient client, SendMessageRequest request, int i)
{
var stopwatch = new Stopwatch();
stopwatch.Start();
var response = await client.SendMessageAsync(request);
stopwatch.Stop();
//Console.WriteLine("&&&#####" + i + "Elapsed ms:" + stopwatch.ElapsedMilliseconds + ". For message ID '" +
// response.MessageId + "':");
i++;
return stopwatch.ElapsedMilliseconds;
}
}

Answer 1

SQS isn't behaving differently in relation to your lambda size, the code which you're running in Lambda is changing it's performance based on the memory/CPU allotted.