| Thanatos 2004-05-30, 11:11 am |
| Hi Mike,
Actually you shouldn't say "san or ethernet"
A storage area network ("SAN") is a description of a topology - a SAN is a
seperate network infrastructure dedicated to storage technology. A SAN can
be implemented using Fibre Channel, or Ethernet, or Token Ring, or ATM or
whatever physical network layer you feel like using. They all have
advantages and disadvantages. In practice, most SAN's are based on a Fibre
Channel physical layer, however Ethernet is gaining ground for SAN's thanks
to the ratification of the iSCSI standard. NetApp and EMC customers have
been building SAN's using NFS protocol on Ethernet (for supporting Oracle
databases and web farms) for over half a decade now.
Ethernet is a physical layer network technology. In fact, for building SAN's
it has some significant advantages over Fibre Channel - it is a true
routable network technology which can carry many different data access
protocols and for which interoperability is guaranteed - lack of
interoperability is a major problem in Fibre Channel environments hence the
publication of a "compatibility matrix" for each vendor. Fibre Channel is
NOT a good networking physical layer, it's just a point-to-point fabric. It
isn't routable (except with McData's iFCP hardware which needs Ethernet in
between to do the routing!) and doesn't handle broadcasts or multicasts.
The big problems for Ethernet as a storage networking technology in the 90's
were lack of performance (100mbit vs. gigabit), which has now turned the
other way as 10Gbit Ethernet has been around for nearly 3 years now and
10Gbit is only now just starting to happen now for Fibre Channel, and CPU
overhead of IP - which is no longer an issue due to offload engines in NIC's
and iSCSI HBAs - an iSCSI HBA looks just like a Fibre Channel HBA or
native SCSI HBA to your OS. With Ethernet based SAN's you get to choose
whether to use a block protocol (iSCSI or iFCP) or file protocol (NFS or
CIFS ) depending on your application requirements.
i.e. I build a clustered SQL server environment last year that used iSCSI
for databases with large fixed records (i.e. like GIS data), and CIFS for
databases with small variable sized transactional records (i.e. financials).
Your right that you will benefit from building a dedicated backup network.
Gigabit Ethernet will be significantly cheaper than even 1Gbit Fibre Channel
for this (particulary where switch ports are concerned), but it won't give
you the raw performance of 2Gbit or 4Gbit Fibre Channel. 10Gbit Ethernet is
still cost probihitive (Intel 10Gbit NIC's are still ~US$3000 and switch
ports from Extreme or Foundry are very expensive) and your server's won't
keep it busy anyway. So your choices are really GbE (for low cost) or 2Gbit
Fibre Channel (raw performance), but the dedicated topology remains the
same.
"Mike" <mikee@mikee.ath.cx> wrote in message
news:107lqhpq9be5nc5@corp.supernews.com...
> One thing that has been recently realized is the backup
> is saturating the primary network (suprise!). I've been
> tasked with finding a solution. I've been researching
> lan-free and server-free backup solutions using a
> SAN. I realized today that even a SAN will eventuall have
> bandwidth problems and as a SAN, can only be used for data
> (tape, shark, etc). I've been thinking going back to a
> dedicated backup network that can be used for backups,
> backups can be scheduled ltos of times, and can also be
> used for server-to-server communications.
>
> Thoughts?
>
> Mike
|