Table 3 - Online Dump Of Active Database With Sql Server; Table 4 - Online Dump Of Active Database With Arcserve - Compaq 219700-001 - ProLiant - 1500 White Paper

Compaq Backup and Recovery for Microsoft SQL Server 6.x
inaccessible, although it is actively accessed only in the first scenario. The two environments do
however, result in quite different performance dynamics.
The system that was used for this testing was a Proliant 5000 server with two (2) Pentium Pro/166
processors and 1 GB of RAM. The system had enough memory and spindles (disk drives) to allow for
a maximum transactional throughput bound by the CPU's. Two 15/30-GB DLT drives were attached
to the system's integrated Fast-Wide SCSI-2 controller. Windows NT 4.0 and SQL Server 6.5 were
installed on the system and a large database was created and loaded.
The SQL Server SMP Stat and Boost Priority parameters were both deactivated (set to 0) for these
tests. Enabling either of these parameters increases the priority of SQL Server threads relative to other
process's threads under Windows NT. We have found that performing online backups with user
activity can cause CPU utilization to rise significantly, and increasing SQL Server priority can increase
the risk of a system 'lock-up' should all CPU's reach 100% usage. Setting SMP Stat to 0 or 'n-1'
(where n = # of CPU's) is especially important when using a third party application (such as
ARCserve) to assist in the backup, or else the SQL Server threads can starve the backup application
and halt the backup process
used if online backups with user activity are to be done.
Table 3 and Table 4 display the results of dumping a database that is under user activity using SQL
Server and ARCserve, respectively. The database was under enough activity to generate a load that is
rd
1/3 (33%) of the system's maximum

Table 3 - Online Dump of Active Database with SQL Server

Transactions Before Dump
Transaction
Rate:
No User
Transactions
33% of
Maximum
Transactions Before Dump
Transaction
Rate:
No User
Transactions
33% of
Maximum
82
Setting SMP Stat to 0 still allows SQL Server threads to use all CPU's present in the system, but guarantees
that one CPU will be available for other processes by moving SQL threads off that CPU if necessary. Refer to
the SQL Server Administrator's Companion for more information on this parameter.
83
We do not recommend attempting a backup of a database that is under a much heavier user read/write load.
¤
1997 Compaq Computer Corporation, All Rights Reserved
82
. Finally, we recommend that an SMP system with at least two CPU's be
83
.
Response CPU Transaction
Time: Usage: Rate:
n/a ~ 0% No User
Transactions
.12 sec ~32% 4% to 16%
of
Maximum

Table 4 - Online Dump of Active Database with ARCserve

Response CPU Transaction
Time: Usage: Rate:
n/a ~ 0% No User
Transactions
.12 sec ~32% 3% to 12%
of
Maximum
Transactions During Dump
Response CPU
Time: Usage:
n/a ~ 6%
1.42 sec to ~64%
0.65 sec
Transactions During Dump
Response CPU
Time: Usage:
n/a ~60%
1.91 sec to ~99%
0.87 sec
Page 63
Dump
Throughput:
9.1 GB/hour
8.9 GB/hour
Dump
Throughput:
10.0 GB/hour
10.0 GB/hour
Doc No 444A/0797