ONTAP Discussions

VM Snapshot and performances

CLOCATEL75
2,256 Views

With a FAS2552 (10 GbE), Data Ontap 8.2.2P1, NFS, on a 17x SAS 600 GB aggregate, a Windows 2008 R2 virtual machine shows low performances with SQLIO random writes (and reads). We applied VSC recommanded settings on ESXi (5.5).

 

workload : Random Write 8K block without cache on 10 GB file

When there is a snapshot on the VM :

 

D:\PARAM\sqlio -kW -s30 -frandom -o8 -b8 -LS -BN -Fparam.txt
sqlio v1.5.SG using system counter for latency timings, 14318180 counts per second parameter file used: param.txt file D:\testfile.dat with 4 threads (0-3) using mask 0x0 (0) 4 threads writing for 30 secs to file D:\testfile.dat using 8KB random IOs enabling multiple I/Os per thread with 8 outstanding buffering set to not use file nor disk caches (as is SQL Server) using specified size: 5000 MB for file: D:\testfile.dat initialization done
CUMULATIVE DATA:
throughput metrics:
IOs/sec: 1762.79
MBs/sec: 13.77
latency metrics:
Min_Latency(ms): 0
Avg_Latency(ms): 17
Max_Latency(ms): 468
histogram:
ms: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24+
%: 0 0 0 0 0 1 2 37 40 5 3 2 1 1 1 1 0 0 0 0 0 0 0 0 5

 

A 10 GB file should be entirely on the cache of the FAS2552. Even when it reads in a VM snapshot.

 

After removing snapshot :

 

f:\PARAM\sqlio -kW -s60 -frandom -o8 -b8 -LS -BN -Fparam.txt
sqlio v1.5.SG using system counter for latency timings, 14318180 counts per second parameter file used: param.txt file F:\testfile.dat with 4 threads (0-3) using mask 0x0 (0) 4 threads writing for 60 secs to file F:\testfile.dat using 8KB random IOs enabling multiple I/Os per thread with 8 outstanding buffering set to not use file nor disk caches (as is SQL Server) using specified size: 10000 MB for file: F:\testfile.dat initialization done
CUMULATIVE DATA:
throughput metrics:
IOs/sec: 15203.99
MBs/sec: 118.78
latency metrics:
Min_Latency(ms): 0
Avg_Latency(ms): 1
Max_Latency(ms): 30
histogram:
ms: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24+
%: 1 63 25 7 3 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

 

Near 10x IOPS.

The difference is a big difference. It's more a VMware subject but it is good to know.

1 REPLY 1
Public