Max Degree of Parallelism, Timeouts and Deadlocks

Question

We have been trying to track down some timeout issues with our SQL Server. Our SQL Server runs a AccPac Account System, an Internal Intranet site and some SharePoint Content DBs.  
We know that some of the queries are complex and take some time, though the timeouts seem to happen randomly and not always on the same pages/queries.   IIS and the SQL timeouts are set to 60 and 120 seconds. 
Looking at some of the SQL Server settings, I noticed that MAXDOP was set to 1. Before doing extensive research on it and looking through the BOL, another DBA and I changed it to 0. 
Server is a Hyper-V VM with:
Processors:4
NUMA nodes:2
Sockets: 2
Interesting thing happened. Our Timeouts seem to have disappeared. Going from several a day, we are now at 1 every few days. Though now the issue we are having is that our Deadlocks have gone through the roof. Going from one or two every few days, we are
up to 8+ a day!
We have been changing our Select statements to include WITH (NOLOCK) so they do not compete with the UPDATE statements they usually fall victim to. The Deadlocks and timeouts do not seem to be related to any of the SharePoint Content DBs. All of the deadlocks
are with our Intranet Site and When it communicates with the AccPac DB, or the Internet Site on its own.
Any Suggestions on where I should be focusing my energy on benchmarking and tuning the server?
Thank you,
  Scott<-

Answer 1

Thank you all for your replies.
The server had 30GB of RAM and then we bumped it up to 40GB at the same time we changed the MAXDOP to 0.
It was set to 1 because if it isn't MS won't support your SharePoint installation. This is from the Setup Guide for SharePoint on SQL Server, MAXDOP = 1 is a must in their book for official support. It always forces serial plans, because to be honest
the sharepoint queries are extremely terrible.
I understand this, though I would guess that the install of SharePoint didn’t actually do the MAXDOP =1 setting during the install? We basically have two Sharepoint sites on the server. One has about 10 users and the other has maybe 20 users. The Sites are
not used very much either. So I didn't think there would be too much impact. 
Though now the issue we are having is that our Deadlocks have gone through the roof.
You probably didn't get this before (though they probably still happened) because the executions were forced serially and you dodged many-a-bullet because of this artificial speed bump. Deadlocks are application based, pure and simple.
The accounting system we are running is something we typically do not alter the DB Contents directly, we are only peering into the database to present information to the user. We looked at READ_COMMITTED_SNAPSHOT, though since that is a DB setting on not
on individual queries, we do not want to alter the Accounting DB as we could not know the potential ramifications of what that would cause.
A Typical Deadlock will occur when the Accounting system is creating or modifying an order’s master record so no one else can modify it, though instead of row lock, it locks the table. This is something that is out of our control. When we do a select against
the same table from the Intranet site, we get a deadlock unless we use the WITH (NOLOCK). The data that we get is not Super Critical. The only potential issue that might happen is that in an uncommitted Transaction form the Accounting system, it could be adding
multiple rows to an order and when we SELECT the data, we might miss a line Item or two.
We have been changing our Select statements to include WITH (NOLOCK) so they do not compete with the UPDATE statements they usually fall victim to.
This really isn't going to get very far to be honest. Are you deadlocking on the same rows? It seems to be the order of operations taken by either the queries or the logic used in them. Without deadlock information there really is nothing to diagnose, but
that definitely sounds like the same resource being used in multiple places when it probably doesn't need to be.
This is one of the typical deadlocks that we get. Intranet Site is getting Totals for Orders while the accounting system is in the process of setting its internal record Lock on an Order. 
<EVENT_INSTANCE>
<EventType>DEADLOCK_GRAPH</EventType>
<PostTime>2014-05-12T15:26:09.447</PostTime>
<SPID>23</SPID>
<TextData>
<deadlock-list>
<deadlock victim="process2f848b048">
<process-list>
<process id="process2f848b048" taskpriority="0" logused="0" waitresource="OBJECT: 12:644249400:0 " waittime="1295" ownerId="247639995" transactionname="SELECT" lasttranstarted="2014-05-12T15:26:08.150" XDES="0x69d1d3620" lockMode="IS" schedulerid="2" kpid="2856" status="suspended" spid="184" sbid="0" ecid="0" priority="0" trancount="0" lastbatchstarted="2014-05-12T15:26:08.150" lastbatchcompleted="2014-05-12T15:26:08.150" lastattention="2014-05-12T14:50:52.280" clientapp=".Net SqlClient Data Provider" hostname="VSVR-WWW-INT12" hostpid="15060" loginname="SFA" isolationlevel="read committed (2)" xactid="247639995" currentdb="7" lockTimeout="4294967295" clientoption1="673185824" clientoption2="128056">
<executionStack>
<frame procname="SFA.dbo.SAGE_SO_order_total_no_history_credit" line="20" stmtstart="1190" stmtend="5542" sqlhandle="0x030007004c642a7a7c17d000e9a100000100000000000000">
SELECT SUM(t.price * t.qtyord) AS On_Order_Total
FROM PRODATA01..somast m
INNER JOIN PRODATA01..sotran t ON m.sono = t.sono
INNER JOIN SFA..item i ON i.our_part_number COLLATE DATABASE_DEFAULT = t.item COLLATE DATABASE_DEFAULT
INNER JOIN SFA..supplier s ON s.supplier_key = i.supplier_key
INNER JOIN SFA..customer c ON c.abbreviation COLLATE DATABASE_DEFAULT = m.custno COLLATE DATABASE_DEFAULT
INNER JOIN SFA..sales_order_ownership soo ON soo.so_id_col = m.id_col
LEFT JOIN PRODATA01..potran p ON p.id_col = t.po_id_col
LEFT JOIN SFA..alloc_inv a ON a.sono COLLATE DATABASE_DEFAULT = t.sono COLLATE DATABASE_DEFAULT AND a.tranlineno = t.tranlineno
WHERE c.is_visible = 1 AND m.sostat NOT IN ('V','X') AND m.sotype IN ('C','','O')
AND t.sostat NOT IN ('V','X') AND t.sotype IN ('C','','O')
--AND t.rqdate BETWEEN @start_ordate AND @end_ordate
AND UPPER(LEFT(t.item,4)) &lt;&gt; 'SHIP' AND t.item NOT LIKE '[_]%'
AND ((SUBSTRING(m.ornum,2,1) = 'A' AND p.expdate &lt;= @end_ordate) OR (t.rqdate &lt;= @en </frame>
</executionStack>
<inputbuf>
Proc [Database Id = 7 Object Id = 2049598540] </inputbuf>
</process>
<process id="process51df0c2c8" taskpriority="0" logused="28364" waitresource="OBJECT: 12:1369823992:0 " waittime="1032" ownerId="247639856" transactionname="user_transaction" lasttranstarted="2014-05-12T15:26:07.940" XDES="0xf8b5b620" lockMode="X" schedulerid="1" kpid="7640" status="suspended" spid="292" sbid="0" ecid="0" priority="0" trancount="2" lastbatchstarted="2014-05-12T15:26:08.410" lastbatchcompleted="2014-05-12T15:26:08.410" clientapp="Sage Pro ERP version 7.5" hostname="VSVR-DESKTOP" hostpid="15892" loginname="AISSCH" isolationlevel="read uncommitted (1)" xactid="247639856" currentdb="12" lockTimeout="4294967295" clientoption1="536870944" clientoption2="128056">
<executionStack>
<frame procname="adhoc" line="1" stmtstart="22" sqlhandle="0x02000000304ac5350b86da8b9422b389413bf23015ac25d0">
UPDATE PRODATA01..SOTRAN WITH (TABLOCK HOLDLOCK) SET lckuser = lckuser WHERE id_col = @P1 </frame>
<frame procname="unknown" line="1" sqlhandle="0x000000000000000000000000000000000000000000000000">
unknown </frame>
</executionStack>
<inputbuf>
(@P1 float)UPDATE PRODATA01..SOTRAN WITH (TABLOCK HOLDLOCK) SET lckuser = lckuser WHERE id_col = @P1 </inputbuf>
</process>
</process-list>
<resource-list>
<objectlock lockPartition="0" objid="644249400" subresource="FULL" dbid="12" objectname="PRODATA01.dbo.somast" id="lock18b5e3900" mode="X" associatedObjectId="644249400">
<owner-list>
<owner id="process51df0c2c8" mode="X" />
</owner-list>
<waiter-list>
<waiter id="process2f848b048" mode="IS" requestType="wait" />
</waiter-list>
</objectlock>
<objectlock lockPartition="0" objid="1369823992" subresource="FULL" dbid="12" objectname="PRODATA01.dbo.sotran" id="lock2ce1c4680" mode="SIX" associatedObjectId="1369823992">
<owner-list>
<owner id="process2f848b048" mode="IS" />
</owner-list>
<waiter-list>
<waiter id="process51df0c2c8" mode="X" requestType="convert" />
</waiter-list>
</objectlock>
</resource-list>
</deadlock>
</deadlock-list>
</TextData>
<TransactionID />
<LoginName>sa</LoginName>
<StartTime>2014-05-12T15:26:09.447</StartTime>
<ServerName>VSVR-SQL</ServerName>
<LoginSid>AQ==</LoginSid>
<EventSequence>2335848</EventSequence>
<IsSystem>1</IsSystem>
<SessionLoginName />
</EVENT_INSTANCE>
I'd (in parallel) look at why parallel plans are being chosen. Not that parallel plans are a bad thing, but is the cost of the execution so high that parallelism is chosen all of the time?
How can I determine the cost of different statements? The Current Cost threshold value is 5.
The last place I would set my effort is on the Dev team. Internal Intranet queries should not take 60 to 120 seconds. That's just asking for issues that you already have. If some larger functionality with that is needed, do it on the back end as
aggregation over a certain time period and use that new static data. Recompute as needed. This is especially true if your deadlocks are happening on these resources (chances are, it is).
We are working with the long queries.  Trying to Break them up.  We thought about backend processing of the data so its available for the user when they need it, but some of the pages that take time are not accessed that often, so if we gathered
the data every 10 minutes in the background it would be called way many more times in a day than would be called on demand. 
Thank you all again!