FC Channel - Capacity

[kartikgajjar]

Hi,

Server has 2 "HP AD193-60001 4Gb Fibre Channel" ports and glance advisorty output is showing following.

BYHBA_AVG_SERVICE_TIME|
BYHBA_REQUEST_QUEUE| |
BYHBA_NAME BYHBA_UTIL| | |
fcd1 96.2 1.0 8.15
fcd0 96.0 1.0 10.19
fcd1 96.0 1.0 8.97
fcd0 96.2 1.0 8.66
fcd1 96.2 1.0 7.82
fcd0 100.0 1.0 8.77
fcd1 100.0 1.0 8.11
fcd0 96.2 1.0 10.00
fcd1 96.0 1.0 8.55
fcd0 96.6 1.0 9.08
fcd1 96.4 1.0 7.99
fcd0 96.4 1.0 9.86
fcd1 96.2 1.0 8.59
fcd0 100.0 1.0 11.09
fcd1 100.0 1.0 9.68
fcd0 96.2 1.0 10.04


Does this mean the FC is fully loaded and we need to add one more FC ?

If Utilization is 100% how come BYHBA_REQUEST_QUEUE is always 1 ?

[Don Mallory]

I hate answering things this way, but the answer is, "it depends".

What is the BYDSK_AVG_SERVICE_TIME of your disks? If it is <= the seek time of your disks, then you will not get a better rate anyways, unless you significantly increase the cache level of your array. The number of pipes to the array won't help.

You can check the GBL_DISK_SUBSYSTEM_QUEUE and also look at the individual disk queues. But if you are seeing a single queued request, and no others queuing up, then your requests are being met as fast as they can be requested. If your request queue is less than 1, then you have HBAs and/or disks that are not in use, which would be a waste of resources.

I would also check the GBL_MEM_CACHE_HIT_PCT value. If your cache hit value is very high, the responses are by and large coming from host cache instead of from the disk anyways.

The most important question here though is... are you finding that the application that is running is having performance issues?

If so, have you also examined other areas of the host other than disk? What about from the array perspective? Are there any particular LVs or PVs that are being hit harder than others? Is it by one specific process?

If not, then your system is responding well. A host can have a high load average, and high level of queue requests, but still be performing just fine if the users are not experiencing issues. When it comes right down to it, the scream test is sometimes the best performance indicator. Everything else is just a way to dig into the actual issue.

Here are a list of some of my favourite stats for generally reviewing application / Database hosts:

GBL_STATDATE
GBL_STATTIME
GBL_CPU_TOTAL_UTIL
GBL_CPU_SYS_MODE_UTIL
GBL_CPU_USER_MODE_UTIL
GBL_MEM_UTIL
GBL_MEM_USER_UTIL
GBL_MEM_SYS_AND_CACHE_UTIL
GBL_MEM_PAGEOUT_RATE
GBL_SWAP_SPACE_UTIL
GBL_MEM_CACHE_HIT_PCT
GBL_MEM_UTIL
GBL_MEM_USER_UTIL
GBL_MEM_SYS_AND_CACHE_UTIL
GBL_MEM_PAGE_REQUEST_RATE
GBL_MEM_PAGEOUT_RATE
GBL_MEM_SWAPOUT_RATE
GBL_SWAP_SPACE_UTIL
GBL_MEM_CACHE_HIT_PCT
GBL_DISK_UTIL_PEAK
GBL_DISK_PHYS_IO_RATE
GBL_DISK_LOGL_IO_RATE
GBL_NET_IN_PACKET_RATE
GBL_NET_OUT_PACKET_RATE
GBL_PRI_QUEUE
GBL_DISK_SUBSYSTEM_QUEUE
GBL_MEM_QUEUE
GBL_NETWORK_SUBSYSTEM_QUEUE
GBL_IPC_SUBSYSTEM_QUEUE
GBL_TERM_IO_QUEUE
GBL_SLEEP_QUEUE
GBL_OTHER_QUEUE
LV_READ_RATE_lvol5
BYDSK_AVG_SERVICE_TIME_1
BYDSK_REQUEST_QUEUE_1

Best regards,
Don

[kartikgajjar]

Thanks for the reply..

will gather more information.

Now, the FC card is "HP AD193-60001 4Gb Fibre Channel"

Setup : FC > Switch > EMC-SAN

I'm not totally into configuration however I believe FC communication has frame size.

I've one server which does small transaction and the other which is more of data warehouse.

Is there anyway to find optimal frame size.