Simple, Reliable Performance for iSCSI Connectivity

In real-world implementations, typical
IOPS performance is much lower than
these benchmarks and is limited by a
number of factors, including application
threading, as well as storage target
processor and disk speeds. At these
lower IOPS levels, processor utilization
is typically quite low, thanks in part
to recent increases in processor
performance.
In addition, Intel Ethernet Server Adapter
assists such as receive-side coalescing and
TCP segmentation ofoad help reduce
processor utilization. Platform features
including receive-side scaling and MSI-X
help accelerate throughput.
By contrast, solutions based on
proprietary ofoad engines may show
little if any improvement in IOPS on
new hardware platforms (although
processor utilization may improve).
This characteristic reects the inherent
limitations to throughput placed on
the overall solution by proprietary
ofoad engines.
In addition to the outstanding IOPS results
shown in Figure 1, the results in Figure
2 show full 10GbE line-rate throughput
at I/O sizes of 8 KB to 64 KB, which
correspond to the most prevalent real-
world cases.
1
At these I/O sizes, processor
utilization falls to relatively low levels.
Figure 2. iSCSI read/write performance on the Intel® 82599 10 Gigabit Ethernet Controller, expressed in terms of throughput.
1
8K4K2K 32K16K 64K 128K 256K 51 2K 1M1K512
0
2
4
6
8
10
12
14
16
18
20
Throughput (Gbps)
I/O Size (Bytes)
Intel® 82599 10 Gigabit Ethernet Controller Single Port Read/Write Throughput Performance
6
Simple, Reliable Performance for iSCSI Connectivity
6