Bit of advice please

[jay_oasis]

Hello

I am looking at setting up a new site with a 100Mbps leased line linking back to HQ

Currently in HQ we have a single SQL server with about 600GB of liive databases on it (I know its a disaster waiting to happen)

What I am looking at doing is setting up a SAN to hold these live Databases and give people in HQ a central file store as well, I would also like to have a second SAN in the new site with instant replication between the SANs in both directions. This would then mean we are DR for each other along with the tape backups in each site and our data is always in sync.

I am looking at about 6TB in total across as many spindals as possible to give us some decent IOPs and some space to grow. If this is a success I can this this happening in a few other sites as well.

I did a quick SQLIO test on our current server, not sure if the file was too small though

C:\SQLIO>call sqlio -kW -s360 -dD -frandom -o8 -b8 -LS -Fparam.txt
sqlio v1.5.SG
using system counter for latency timings, -1302387296 counts per second
parameter file used: param.txt
file d:\testfile.dat with 2 threads (0-1) using mask 0x0 (0)
2 threads writing for 360 secs to file d:\testfile.dat
using 8KB random IOs
enabling multiple I/Os per thread with 8 outstanding
using specified size: 100 MB for file: d:\testfile.dat
initialization done
CUMULATIVE DATA:
throughput metrics:
IOs/sec: 38815.81
MBs/sec: 303.24
latency metrics:
Min_Latency(ms): 0
Avg_Latency(ms): 0
Max_Latency(ms): 2967
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+
%: 100 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

Can anyone give me a little bit of advice?

[Johan Guldmyr]

Hi, 38000 IOPS is a lot. Especially with spinning disks. I'd say use a file that's larger than RAM, so that it's not cached. Also if you have a 100MB/s file and the speed is measured to be 303MB/s, you need something bigger to get a test that runs for longer than 1/3 of a second.

HP has the EVA4400 or the new P6000 that does both asynchronous and synchronous replications via 'continuous access'. The P2000 also has replication but historically the P2000 has had a lot of problems. It appears to be more stable these days but IMO the replication mechanisms/tools are better on the EVA.

Also the replication with the EVA tends to want all bandwidth that's available.

Are you planning to use the leased line for other traffic? Can you run native fibre channel or do you want to run something else like iSCSI or FCIP?

Maybe the lefthand/3PAR gear has replication too.

[jay_oasis]

Thanks for the info

I thought that that file was too small, I will have to wait for a quiet slot on the server to run the test again.

I always hear different things, some say iSCSI is best other say FC is best, what do you think?

What happens to files that are in use when replicated? eg what if we want to replicate databases while they are in use on both ends?

Oh and yes we will be using the leased line for other things, mainly some Citrix hosting (about 30 users) and internet for about 10 people (mainly very light) at the new office and we have 200 users and Citrix at HQ, I have never seen usage go above a consistant 20Mbps.

Thanks

[Johan Guldmyr]

I thought that that file was too small, I will have to wait for a quiet slot on the server to run the test again.

I always hear different things, some say iSCSI is best other say FC is best, what do you think?

>>> I would be pro-FC but I have little experience with iSCSI. iSCSI is not lossless as it runs over TCP/IP. FC generally is. iSCSI would have longer response times but as you have some kind of leased line - maybe you need to run traffic over TCP/IP anyway? iSCSI is generally much cheaper.

What happens to files that are in use when replicated? eg what if we want to replicate databases while they are in use on both ends?

>>> For reporting/backup you could make a copy of the database/LUN on the array. Check out snapshots.

As for a database for example that is in use on both ends, you would need to use synchronous replication and of course a cluster. If asynchronous replication is employed and the LUNs are used on both sides that makes it difficult, because one side would always be lagging. But, maybe there are solutions around this as well.

Oh and yes we will be using the leased line for other things, mainly some Citrix hosting (about 30 users) and internet for about 10 people (mainly very light) at the new office and we have 200 users and Citrix at HQ, I have never seen usage go above a consistant 20Mbps.

>>> I'd consider limiting / getting two separate lines. So that "other stuff" cannot go over 20Mbps and that you then have the 80Mbps dedicated for replication, it's especially handy if you need to re-initialize the replication (copy everything from source to destination). Especially if you run synchronous replication on the EVA it's important to not share the line. Maybe other replication solutions can handle varying bandwidth in a better way.

With synchronous replication the hosts aren't told that the write is completed until it's been written on both sites. So if you have a line with high response time it can cause big problems.