Linq2SQL vs NHibernate performance (have I gone mad?)

Question

I have written the following tests to compare performance of Linq2SQL and NHibernate and I find results to be somewhat strange. Mappings are straight forward and identical for both. Both are running against a live DB. Although I'm not deleting Campaigns in case of Linq, but that shouldn't affect performance by more than 10 ms.

Linq:

    [Test]
    public void Test1000ReadsWritesToAgentStateLinqPrecompiled()
    {
        Stopwatch sw = new Stopwatch();
        Stopwatch swIn = new Stopwatch();
        sw.Start();
        for (int i = 0; i < 1000; i++)
        {
            swIn.Reset();
            swIn.Start();
            ReadWriteAndDeleteAgentStateWithLinqPrecompiled();
            swIn.Stop();
            Console.WriteLine("Run ReadWriteAndDeleteAgentState: " + swIn.ElapsedMilliseconds + " ms");
        }

        sw.Stop();
        Console.WriteLine("Total Time: " + sw.ElapsedMilliseconds + " ms");
        Console.WriteLine("Average time to execute queries: " + sw.ElapsedMilliseconds / 1000 + " ms");
    }

    private static readonly Func<AgentDesktop3DataContext, int, EntityModel.CampaignDetail>
        GetCampaignById =
            CompiledQuery.Compile<AgentDesktop3DataContext, int, EntityModel.CampaignDetail>(
                (ctx, sessionId) => (from cd in ctx.CampaignDetails
                                     join a in ctx.AgentCampaigns on cd.CampaignDetailId equals a.CampaignDetailId
                                     where a.AgentStateId == sessionId
                                     select cd).FirstOrDefault());

    private void ReadWriteAndDeleteAgentStateWithLinqPrecompiled()
    {
        int id = 0;
        using (var ctx = new AgentDesktop3DataContext())
        {
            EntityModel.AgentState agentState = new EntityModel.AgentState();
            var campaign = new EntityModel.CampaignDetail { CampaignName = "Test" };
            var campaignDisposition = new EntityModel.CampaignDisposition { Code = "123" };
            campaignDisposition.Description = "abc";
            campaign.CampaignDispositions.Add(campaignDisposition);
            agentState.CallState = 3;

            campaign.AgentCampaigns.Add(new AgentCampaign
            {
                AgentState = agentState
            });
            ctx.CampaignDetails.InsertOnSubmit(campaign);
            ctx.AgentStates.InsertOnSubmit(agentState);
            ctx.SubmitChanges();
            id = agentState.AgentStateId;
        }

        using (var ctx = new AgentDesktop3DataContext())
        {
            var dbAgentState = ctx.GetAgentStateById(id);
            Assert.IsNotNull(dbAgentState);
            Assert.AreEqual(dbAgentState.CallState, 3);
            var campaignDetails = GetCampaignById(ctx, id);

            Assert.AreEqual(campaignDetails.CampaignDispositions[0].Description, "abc");
        }

        using (var ctx = new AgentDesktop3DataContext())
        {
            ctx.DeleteSessionById(id);
        }
    }

NHibernate (the loop is the same):

    private void ReadWriteAndDeleteAgentState()
    {
            var id = WriteAgentState().Id;
            StartNewTransaction();

            var dbAgentState = agentStateRepository.Get(id);

            Assert.IsNotNull(dbAgentState);
            Assert.AreEqual(dbAgentState.CallState, 3);
            Assert.AreEqual(dbAgentState.Campaigns[0].Dispositions[0].Description, "abc");

            var campaignId = dbAgentState.Campaigns[0].Id;
            agentStateRepository.Delete(dbAgentState);

            NHibernateSession.Current.Transaction.Commit();

            Cleanup(campaignId);

            NHibernateSession.Current.BeginTransaction();
    }

Results:

NHibernate:
Total Time: 9469 ms
Average time to execute 13 queries: 9 ms

Linq:
Total Time: 127200 ms
Average time to execute 13 queries: 127 ms

Linq lost by 13.5 times! Event with precompiled queries (both read queries are precompiled).

This can't be right, although I expected NHibernate to be faster, this is just too big of a difference, considering mappings are identical and NHibernate actually executes more queries against the DB.

Update. I have refactored a project to use NHibernate instead of Linq2Sql and the performance gain seems to be a lot less (about 20-30%) compared to test working on the same mappings. Does anyone have some real world examples of their own?

Answer 1

Why are you surprised? People which know ORM's (or write one themselves) in generally think LING (and EF) are not exactly "top of the line" Products at this moment ;) 3rd grade attempt. Maybe ;)

Answer 2

Run a profiler, both on the .NET code and on the SQL Server database. Also, identify what SQL statements are being run under the covers for both scenarios. Where is the time being lost for LinqToSql? If the underlying SQL statements are different, why? It's very likely you can tweak both ORMs to be faster. They should likely be in the same ballpark performance wise for simple tests. This feels like a configuration problem.